1 /* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <[email protected]> 6 * Florian La Roche, <[email protected]> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #ifndef _LINUX_SKBUFF_H 15 #define _LINUX_SKBUFF_H 16 17 #include <linux/kernel.h> 18 #include <linux/kmemcheck.h> 19 #include <linux/compiler.h> 20 #include <linux/time.h> 21 #include <linux/bug.h> 22 #include <linux/cache.h> 23 #include <linux/rbtree.h> 24 #include <linux/socket.h> 25 26 #include <linux/atomic.h> 27 #include <asm/types.h> 28 #include <linux/spinlock.h> 29 #include <linux/net.h> 30 #include <linux/textsearch.h> 31 #include <net/checksum.h> 32 #include <linux/rcupdate.h> 33 #include <linux/hrtimer.h> 34 #include <linux/dma-mapping.h> 35 #include <linux/netdev_features.h> 36 #include <linux/sched.h> 37 #include <net/flow_dissector.h> 38 #include <linux/splice.h> 39 #include <linux/in6.h> 40 #include <linux/if_packet.h> 41 #include <net/flow.h> 42 43 /* The interface for checksum offload between the stack and networking drivers 44 * is as follows... 45 * 46 * A. IP checksum related features 47 * 48 * Drivers advertise checksum offload capabilities in the features of a device. 49 * From the stack's point of view these are capabilities offered by the driver, 50 * a driver typically only advertises features that it is capable of offloading 51 * to its device. 52 * 53 * The checksum related features are: 54 * 55 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 56 * IP (one's complement) checksum for any combination 57 * of protocols or protocol layering. The checksum is 58 * computed and set in a packet per the CHECKSUM_PARTIAL 59 * interface (see below). 60 * 61 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 62 * TCP or UDP packets over IPv4. These are specifically 63 * unencapsulated packets of the form IPv4|TCP or 64 * IPv4|UDP where the Protocol field in the IPv4 header 65 * is TCP or UDP. The IPv4 header may contain IP options 66 * This feature cannot be set in features for a device 67 * with NETIF_F_HW_CSUM also set. This feature is being 68 * DEPRECATED (see below). 69 * 70 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 71 * TCP or UDP packets over IPv6. These are specifically 72 * unencapsulated packets of the form IPv6|TCP or 73 * IPv4|UDP where the Next Header field in the IPv6 74 * header is either TCP or UDP. IPv6 extension headers 75 * are not supported with this feature. This feature 76 * cannot be set in features for a device with 77 * NETIF_F_HW_CSUM also set. This feature is being 78 * DEPRECATED (see below). 79 * 80 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 81 * This flag is used only used to disable the RX checksum 82 * feature for a device. The stack will accept receive 83 * checksum indication in packets received on a device 84 * regardless of whether NETIF_F_RXCSUM is set. 85 * 86 * B. Checksumming of received packets by device. Indication of checksum 87 * verification is in set skb->ip_summed. Possible values are: 88 * 89 * CHECKSUM_NONE: 90 * 91 * Device did not checksum this packet e.g. due to lack of capabilities. 92 * The packet contains full (though not verified) checksum in packet but 93 * not in skb->csum. Thus, skb->csum is undefined in this case. 94 * 95 * CHECKSUM_UNNECESSARY: 96 * 97 * The hardware you're dealing with doesn't calculate the full checksum 98 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 99 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 100 * if their checksums are okay. skb->csum is still undefined in this case 101 * though. A driver or device must never modify the checksum field in the 102 * packet even if checksum is verified. 103 * 104 * CHECKSUM_UNNECESSARY is applicable to following protocols: 105 * TCP: IPv6 and IPv4. 106 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 107 * zero UDP checksum for either IPv4 or IPv6, the networking stack 108 * may perform further validation in this case. 109 * GRE: only if the checksum is present in the header. 110 * SCTP: indicates the CRC in SCTP header has been validated. 111 * 112 * skb->csum_level indicates the number of consecutive checksums found in 113 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 114 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 115 * and a device is able to verify the checksums for UDP (possibly zero), 116 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 117 * two. If the device were only able to verify the UDP checksum and not 118 * GRE, either because it doesn't support GRE checksum of because GRE 119 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 120 * not considered in this case). 121 * 122 * CHECKSUM_COMPLETE: 123 * 124 * This is the most generic way. The device supplied checksum of the _whole_ 125 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 126 * hardware doesn't need to parse L3/L4 headers to implement this. 127 * 128 * Note: Even if device supports only some protocols, but is able to produce 129 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 130 * 131 * CHECKSUM_PARTIAL: 132 * 133 * A checksum is set up to be offloaded to a device as described in the 134 * output description for CHECKSUM_PARTIAL. This may occur on a packet 135 * received directly from another Linux OS, e.g., a virtualized Linux kernel 136 * on the same host, or it may be set in the input path in GRO or remote 137 * checksum offload. For the purposes of checksum verification, the checksum 138 * referred to by skb->csum_start + skb->csum_offset and any preceding 139 * checksums in the packet are considered verified. Any checksums in the 140 * packet that are after the checksum being offloaded are not considered to 141 * be verified. 142 * 143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 144 * in the skb->ip_summed for a packet. Values are: 145 * 146 * CHECKSUM_PARTIAL: 147 * 148 * The driver is required to checksum the packet as seen by hard_start_xmit() 149 * from skb->csum_start up to the end, and to record/write the checksum at 150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 151 * csum_start and csum_offset values are valid values given the length and 152 * offset of the packet, however they should not attempt to validate that the 153 * checksum refers to a legitimate transport layer checksum-- it is the 154 * purview of the stack to validate that csum_start and csum_offset are set 155 * correctly. 156 * 157 * When the stack requests checksum offload for a packet, the driver MUST 158 * ensure that the checksum is set correctly. A driver can either offload the 159 * checksum calculation to the device, or call skb_checksum_help (in the case 160 * that the device does not support offload for a particular checksum). 161 * 162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 164 * checksum offload capability. If a device has limited checksum capabilities 165 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as 166 * described above) a helper function can be called to resolve 167 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper 168 * function takes a spec argument that describes the protocol layer that is 169 * supported for checksum offload and can be called for each packet. If a 170 * packet does not match the specification for offload, skb_checksum_help 171 * is called to resolve the checksum. 172 * 173 * CHECKSUM_NONE: 174 * 175 * The skb was already checksummed by the protocol, or a checksum is not 176 * required. 177 * 178 * CHECKSUM_UNNECESSARY: 179 * 180 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 181 * output. 182 * 183 * CHECKSUM_COMPLETE: 184 * Not used in checksum output. If a driver observes a packet with this value 185 * set in skbuff, if should treat as CHECKSUM_NONE being set. 186 * 187 * D. Non-IP checksum (CRC) offloads 188 * 189 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 190 * offloading the SCTP CRC in a packet. To perform this offload the stack 191 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 192 * accordingly. Note the there is no indication in the skbuff that the 193 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports 194 * both IP checksum offload and SCTP CRC offload must verify which offload 195 * is configured for a packet presumably by inspecting packet headers. 196 * 197 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 198 * offloading the FCOE CRC in a packet. To perform this offload the stack 199 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 200 * accordingly. Note the there is no indication in the skbuff that the 201 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 202 * both IP checksum offload and FCOE CRC offload must verify which offload 203 * is configured for a packet presumably by inspecting packet headers. 204 * 205 * E. Checksumming on output with GSO. 206 * 207 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 208 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 209 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 210 * part of the GSO operation is implied. If a checksum is being offloaded 211 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 212 * are set to refer to the outermost checksum being offload (two offloaded 213 * checksums are possible with UDP encapsulation). 214 */ 215 216 /* Don't change this without changing skb_csum_unnecessary! */ 217 #define CHECKSUM_NONE 0 218 #define CHECKSUM_UNNECESSARY 1 219 #define CHECKSUM_COMPLETE 2 220 #define CHECKSUM_PARTIAL 3 221 222 /* Maximum value in skb->csum_level */ 223 #define SKB_MAX_CSUM_LEVEL 3 224 225 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 226 #define SKB_WITH_OVERHEAD(X) \ 227 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 228 #define SKB_MAX_ORDER(X, ORDER) \ 229 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 230 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 231 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 232 233 /* return minimum truesize of one skb containing X bytes of data */ 234 #define SKB_TRUESIZE(X) ((X) + \ 235 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 236 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 237 238 struct net_device; 239 struct scatterlist; 240 struct pipe_inode_info; 241 struct iov_iter; 242 struct napi_struct; 243 244 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 245 struct nf_conntrack { 246 atomic_t use; 247 }; 248 #endif 249 250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 251 struct nf_bridge_info { 252 atomic_t use; 253 enum { 254 BRNF_PROTO_UNCHANGED, 255 BRNF_PROTO_8021Q, 256 BRNF_PROTO_PPPOE 257 } orig_proto:8; 258 u8 pkt_otherhost:1; 259 u8 in_prerouting:1; 260 u8 bridged_dnat:1; 261 __u16 frag_max_size; 262 struct net_device *physindev; 263 264 /* always valid & non-NULL from FORWARD on, for physdev match */ 265 struct net_device *physoutdev; 266 union { 267 /* prerouting: detect dnat in orig/reply direction */ 268 __be32 ipv4_daddr; 269 struct in6_addr ipv6_daddr; 270 271 /* after prerouting + nat detected: store original source 272 * mac since neigh resolution overwrites it, only used while 273 * skb is out in neigh layer. 274 */ 275 char neigh_header[8]; 276 }; 277 }; 278 #endif 279 280 struct sk_buff_head { 281 /* These two members must be first. */ 282 struct sk_buff *next; 283 struct sk_buff *prev; 284 285 __u32 qlen; 286 spinlock_t lock; 287 }; 288 289 struct sk_buff; 290 291 /* To allow 64K frame to be packed as single skb without frag_list we 292 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 293 * buffers which do not start on a page boundary. 294 * 295 * Since GRO uses frags we allocate at least 16 regardless of page 296 * size. 297 */ 298 #if (65536/PAGE_SIZE + 1) < 16 299 #define MAX_SKB_FRAGS 16UL 300 #else 301 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 302 #endif 303 extern int sysctl_max_skb_frags; 304 305 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 306 * segment using its current segmentation instead. 307 */ 308 #define GSO_BY_FRAGS 0xFFFF 309 310 typedef struct skb_frag_struct skb_frag_t; 311 312 struct skb_frag_struct { 313 struct { 314 struct page *p; 315 } page; 316 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 317 __u32 page_offset; 318 __u32 size; 319 #else 320 __u16 page_offset; 321 __u16 size; 322 #endif 323 }; 324 325 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 326 { 327 return frag->size; 328 } 329 330 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 331 { 332 frag->size = size; 333 } 334 335 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 336 { 337 frag->size += delta; 338 } 339 340 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 341 { 342 frag->size -= delta; 343 } 344 345 #define HAVE_HW_TIME_STAMP 346 347 /** 348 * struct skb_shared_hwtstamps - hardware time stamps 349 * @hwtstamp: hardware time stamp transformed into duration 350 * since arbitrary point in time 351 * 352 * Software time stamps generated by ktime_get_real() are stored in 353 * skb->tstamp. 354 * 355 * hwtstamps can only be compared against other hwtstamps from 356 * the same device. 357 * 358 * This structure is attached to packets as part of the 359 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 360 */ 361 struct skb_shared_hwtstamps { 362 ktime_t hwtstamp; 363 }; 364 365 /* Definitions for tx_flags in struct skb_shared_info */ 366 enum { 367 /* generate hardware time stamp */ 368 SKBTX_HW_TSTAMP = 1 << 0, 369 370 /* generate software time stamp when queueing packet to NIC */ 371 SKBTX_SW_TSTAMP = 1 << 1, 372 373 /* device driver is going to provide hardware time stamp */ 374 SKBTX_IN_PROGRESS = 1 << 2, 375 376 /* device driver supports TX zero-copy buffers */ 377 SKBTX_DEV_ZEROCOPY = 1 << 3, 378 379 /* generate wifi status information (where possible) */ 380 SKBTX_WIFI_STATUS = 1 << 4, 381 382 /* This indicates at least one fragment might be overwritten 383 * (as in vmsplice(), sendfile() ...) 384 * If we need to compute a TX checksum, we'll need to copy 385 * all frags to avoid possible bad checksum 386 */ 387 SKBTX_SHARED_FRAG = 1 << 5, 388 389 /* generate software time stamp when entering packet scheduling */ 390 SKBTX_SCHED_TSTAMP = 1 << 6, 391 }; 392 393 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 394 SKBTX_SCHED_TSTAMP) 395 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 396 397 /* 398 * The callback notifies userspace to release buffers when skb DMA is done in 399 * lower device, the skb last reference should be 0 when calling this. 400 * The zerocopy_success argument is true if zero copy transmit occurred, 401 * false on data copy or out of memory error caused by data copy attempt. 402 * The ctx field is used to track device context. 403 * The desc field is used to track userspace buffer index. 404 */ 405 struct ubuf_info { 406 void (*callback)(struct ubuf_info *, bool zerocopy_success); 407 void *ctx; 408 unsigned long desc; 409 }; 410 411 /* This data is invariant across clones and lives at 412 * the end of the header data, ie. at skb->end. 413 */ 414 struct skb_shared_info { 415 unsigned char nr_frags; 416 __u8 tx_flags; 417 unsigned short gso_size; 418 /* Warning: this field is not always filled in (UFO)! */ 419 unsigned short gso_segs; 420 unsigned short gso_type; 421 struct sk_buff *frag_list; 422 struct skb_shared_hwtstamps hwtstamps; 423 u32 tskey; 424 __be32 ip6_frag_id; 425 426 /* 427 * Warning : all fields before dataref are cleared in __alloc_skb() 428 */ 429 atomic_t dataref; 430 431 /* Intermediate layers must ensure that destructor_arg 432 * remains valid until skb destructor */ 433 void * destructor_arg; 434 435 /* must be last field, see pskb_expand_head() */ 436 skb_frag_t frags[MAX_SKB_FRAGS]; 437 }; 438 439 /* We divide dataref into two halves. The higher 16 bits hold references 440 * to the payload part of skb->data. The lower 16 bits hold references to 441 * the entire skb->data. A clone of a headerless skb holds the length of 442 * the header in skb->hdr_len. 443 * 444 * All users must obey the rule that the skb->data reference count must be 445 * greater than or equal to the payload reference count. 446 * 447 * Holding a reference to the payload part means that the user does not 448 * care about modifications to the header part of skb->data. 449 */ 450 #define SKB_DATAREF_SHIFT 16 451 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 452 453 454 enum { 455 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 456 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 457 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 458 }; 459 460 enum { 461 SKB_GSO_TCPV4 = 1 << 0, 462 SKB_GSO_UDP = 1 << 1, 463 464 /* This indicates the skb is from an untrusted source. */ 465 SKB_GSO_DODGY = 1 << 2, 466 467 /* This indicates the tcp segment has CWR set. */ 468 SKB_GSO_TCP_ECN = 1 << 3, 469 470 SKB_GSO_TCP_FIXEDID = 1 << 4, 471 472 SKB_GSO_TCPV6 = 1 << 5, 473 474 SKB_GSO_FCOE = 1 << 6, 475 476 SKB_GSO_GRE = 1 << 7, 477 478 SKB_GSO_GRE_CSUM = 1 << 8, 479 480 SKB_GSO_IPXIP4 = 1 << 9, 481 482 SKB_GSO_IPXIP6 = 1 << 10, 483 484 SKB_GSO_UDP_TUNNEL = 1 << 11, 485 486 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12, 487 488 SKB_GSO_PARTIAL = 1 << 13, 489 490 SKB_GSO_TUNNEL_REMCSUM = 1 << 14, 491 492 SKB_GSO_SCTP = 1 << 15, 493 }; 494 495 #if BITS_PER_LONG > 32 496 #define NET_SKBUFF_DATA_USES_OFFSET 1 497 #endif 498 499 #ifdef NET_SKBUFF_DATA_USES_OFFSET 500 typedef unsigned int sk_buff_data_t; 501 #else 502 typedef unsigned char *sk_buff_data_t; 503 #endif 504 505 /** 506 * struct skb_mstamp - multi resolution time stamps 507 * @stamp_us: timestamp in us resolution 508 * @stamp_jiffies: timestamp in jiffies 509 */ 510 struct skb_mstamp { 511 union { 512 u64 v64; 513 struct { 514 u32 stamp_us; 515 u32 stamp_jiffies; 516 }; 517 }; 518 }; 519 520 /** 521 * skb_mstamp_get - get current timestamp 522 * @cl: place to store timestamps 523 */ 524 static inline void skb_mstamp_get(struct skb_mstamp *cl) 525 { 526 u64 val = local_clock(); 527 528 do_div(val, NSEC_PER_USEC); 529 cl->stamp_us = (u32)val; 530 cl->stamp_jiffies = (u32)jiffies; 531 } 532 533 /** 534 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 535 * @t1: pointer to newest sample 536 * @t0: pointer to oldest sample 537 */ 538 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 539 const struct skb_mstamp *t0) 540 { 541 s32 delta_us = t1->stamp_us - t0->stamp_us; 542 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 543 544 /* If delta_us is negative, this might be because interval is too big, 545 * or local_clock() drift is too big : fallback using jiffies. 546 */ 547 if (delta_us <= 0 || 548 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 549 550 delta_us = jiffies_to_usecs(delta_jiffies); 551 552 return delta_us; 553 } 554 555 static inline bool skb_mstamp_after(const struct skb_mstamp *t1, 556 const struct skb_mstamp *t0) 557 { 558 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies; 559 560 if (!diff) 561 diff = t1->stamp_us - t0->stamp_us; 562 return diff > 0; 563 } 564 565 /** 566 * struct sk_buff - socket buffer 567 * @next: Next buffer in list 568 * @prev: Previous buffer in list 569 * @tstamp: Time we arrived/left 570 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 571 * @sk: Socket we are owned by 572 * @dev: Device we arrived on/are leaving by 573 * @cb: Control buffer. Free for use by every layer. Put private vars here 574 * @_skb_refdst: destination entry (with norefcount bit) 575 * @sp: the security path, used for xfrm 576 * @len: Length of actual data 577 * @data_len: Data length 578 * @mac_len: Length of link layer header 579 * @hdr_len: writable header length of cloned skb 580 * @csum: Checksum (must include start/offset pair) 581 * @csum_start: Offset from skb->head where checksumming should start 582 * @csum_offset: Offset from csum_start where checksum should be stored 583 * @priority: Packet queueing priority 584 * @ignore_df: allow local fragmentation 585 * @cloned: Head may be cloned (check refcnt to be sure) 586 * @ip_summed: Driver fed us an IP checksum 587 * @nohdr: Payload reference only, must not modify header 588 * @nfctinfo: Relationship of this skb to the connection 589 * @pkt_type: Packet class 590 * @fclone: skbuff clone status 591 * @ipvs_property: skbuff is owned by ipvs 592 * @tc_skip_classify: do not classify packet. set by IFB device 593 * @tc_at_ingress: used within tc_classify to distinguish in/egress 594 * @tc_redirected: packet was redirected by a tc action 595 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 596 * @peeked: this packet has been seen already, so stats have been 597 * done for it, don't do them again 598 * @nf_trace: netfilter packet trace flag 599 * @protocol: Packet protocol from driver 600 * @destructor: Destruct function 601 * @nfct: Associated connection, if any 602 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 603 * @skb_iif: ifindex of device we arrived on 604 * @tc_index: Traffic control index 605 * @hash: the packet hash 606 * @queue_mapping: Queue mapping for multiqueue devices 607 * @xmit_more: More SKBs are pending for this queue 608 * @ndisc_nodetype: router type (from link layer) 609 * @ooo_okay: allow the mapping of a socket to a queue to be changed 610 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 611 * ports. 612 * @sw_hash: indicates hash was computed in software stack 613 * @wifi_acked_valid: wifi_acked was set 614 * @wifi_acked: whether frame was acked on wifi or not 615 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 616 * @napi_id: id of the NAPI struct this skb came from 617 * @secmark: security marking 618 * @mark: Generic packet mark 619 * @vlan_proto: vlan encapsulation protocol 620 * @vlan_tci: vlan tag control information 621 * @inner_protocol: Protocol (encapsulation) 622 * @inner_transport_header: Inner transport layer header (encapsulation) 623 * @inner_network_header: Network layer header (encapsulation) 624 * @inner_mac_header: Link layer header (encapsulation) 625 * @transport_header: Transport layer header 626 * @network_header: Network layer header 627 * @mac_header: Link layer header 628 * @tail: Tail pointer 629 * @end: End pointer 630 * @head: Head of buffer 631 * @data: Data head pointer 632 * @truesize: Buffer size 633 * @users: User count - see {datagram,tcp}.c 634 */ 635 636 struct sk_buff { 637 union { 638 struct { 639 /* These two members must be first. */ 640 struct sk_buff *next; 641 struct sk_buff *prev; 642 643 union { 644 ktime_t tstamp; 645 struct skb_mstamp skb_mstamp; 646 }; 647 }; 648 struct rb_node rbnode; /* used in netem & tcp stack */ 649 }; 650 struct sock *sk; 651 652 union { 653 struct net_device *dev; 654 /* Some protocols might use this space to store information, 655 * while device pointer would be NULL. 656 * UDP receive path is one user. 657 */ 658 unsigned long dev_scratch; 659 }; 660 /* 661 * This is the control buffer. It is free to use for every 662 * layer. Please put your private variables there. If you 663 * want to keep them across layers you have to do a skb_clone() 664 * first. This is owned by whoever has the skb queued ATM. 665 */ 666 char cb[48] __aligned(8); 667 668 unsigned long _skb_refdst; 669 void (*destructor)(struct sk_buff *skb); 670 #ifdef CONFIG_XFRM 671 struct sec_path *sp; 672 #endif 673 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 674 struct nf_conntrack *nfct; 675 #endif 676 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 677 struct nf_bridge_info *nf_bridge; 678 #endif 679 unsigned int len, 680 data_len; 681 __u16 mac_len, 682 hdr_len; 683 684 /* Following fields are _not_ copied in __copy_skb_header() 685 * Note that queue_mapping is here mostly to fill a hole. 686 */ 687 kmemcheck_bitfield_begin(flags1); 688 __u16 queue_mapping; 689 690 /* if you move cloned around you also must adapt those constants */ 691 #ifdef __BIG_ENDIAN_BITFIELD 692 #define CLONED_MASK (1 << 7) 693 #else 694 #define CLONED_MASK 1 695 #endif 696 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 697 698 __u8 __cloned_offset[0]; 699 __u8 cloned:1, 700 nohdr:1, 701 fclone:2, 702 peeked:1, 703 head_frag:1, 704 xmit_more:1, 705 __unused:1; /* one bit hole */ 706 kmemcheck_bitfield_end(flags1); 707 708 /* fields enclosed in headers_start/headers_end are copied 709 * using a single memcpy() in __copy_skb_header() 710 */ 711 /* private: */ 712 __u32 headers_start[0]; 713 /* public: */ 714 715 /* if you move pkt_type around you also must adapt those constants */ 716 #ifdef __BIG_ENDIAN_BITFIELD 717 #define PKT_TYPE_MAX (7 << 5) 718 #else 719 #define PKT_TYPE_MAX 7 720 #endif 721 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 722 723 __u8 __pkt_type_offset[0]; 724 __u8 pkt_type:3; 725 __u8 pfmemalloc:1; 726 __u8 ignore_df:1; 727 __u8 nfctinfo:3; 728 729 __u8 nf_trace:1; 730 __u8 ip_summed:2; 731 __u8 ooo_okay:1; 732 __u8 l4_hash:1; 733 __u8 sw_hash:1; 734 __u8 wifi_acked_valid:1; 735 __u8 wifi_acked:1; 736 737 __u8 no_fcs:1; 738 /* Indicates the inner headers are valid in the skbuff. */ 739 __u8 encapsulation:1; 740 __u8 encap_hdr_csum:1; 741 __u8 csum_valid:1; 742 __u8 csum_complete_sw:1; 743 __u8 csum_level:2; 744 __u8 csum_bad:1; 745 746 #ifdef CONFIG_IPV6_NDISC_NODETYPE 747 __u8 ndisc_nodetype:2; 748 #endif 749 __u8 ipvs_property:1; 750 __u8 inner_protocol_type:1; 751 __u8 remcsum_offload:1; 752 #ifdef CONFIG_NET_SWITCHDEV 753 __u8 offload_fwd_mark:1; 754 #endif 755 #ifdef CONFIG_NET_CLS_ACT 756 __u8 tc_skip_classify:1; 757 __u8 tc_at_ingress:1; 758 __u8 tc_redirected:1; 759 __u8 tc_from_ingress:1; 760 #endif 761 762 #ifdef CONFIG_NET_SCHED 763 __u16 tc_index; /* traffic control index */ 764 #endif 765 766 union { 767 __wsum csum; 768 struct { 769 __u16 csum_start; 770 __u16 csum_offset; 771 }; 772 }; 773 __u32 priority; 774 int skb_iif; 775 __u32 hash; 776 __be16 vlan_proto; 777 __u16 vlan_tci; 778 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 779 union { 780 unsigned int napi_id; 781 unsigned int sender_cpu; 782 }; 783 #endif 784 #ifdef CONFIG_NETWORK_SECMARK 785 __u32 secmark; 786 #endif 787 788 union { 789 __u32 mark; 790 __u32 reserved_tailroom; 791 }; 792 793 union { 794 __be16 inner_protocol; 795 __u8 inner_ipproto; 796 }; 797 798 __u16 inner_transport_header; 799 __u16 inner_network_header; 800 __u16 inner_mac_header; 801 802 __be16 protocol; 803 __u16 transport_header; 804 __u16 network_header; 805 __u16 mac_header; 806 807 /* private: */ 808 __u32 headers_end[0]; 809 /* public: */ 810 811 /* These elements must be at the end, see alloc_skb() for details. */ 812 sk_buff_data_t tail; 813 sk_buff_data_t end; 814 unsigned char *head, 815 *data; 816 unsigned int truesize; 817 atomic_t users; 818 }; 819 820 #ifdef __KERNEL__ 821 /* 822 * Handling routines are only of interest to the kernel 823 */ 824 #include <linux/slab.h> 825 826 827 #define SKB_ALLOC_FCLONE 0x01 828 #define SKB_ALLOC_RX 0x02 829 #define SKB_ALLOC_NAPI 0x04 830 831 /* Returns true if the skb was allocated from PFMEMALLOC reserves */ 832 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 833 { 834 return unlikely(skb->pfmemalloc); 835 } 836 837 /* 838 * skb might have a dst pointer attached, refcounted or not. 839 * _skb_refdst low order bit is set if refcount was _not_ taken 840 */ 841 #define SKB_DST_NOREF 1UL 842 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 843 844 /** 845 * skb_dst - returns skb dst_entry 846 * @skb: buffer 847 * 848 * Returns skb dst_entry, regardless of reference taken or not. 849 */ 850 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 851 { 852 /* If refdst was not refcounted, check we still are in a 853 * rcu_read_lock section 854 */ 855 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 856 !rcu_read_lock_held() && 857 !rcu_read_lock_bh_held()); 858 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 859 } 860 861 /** 862 * skb_dst_set - sets skb dst 863 * @skb: buffer 864 * @dst: dst entry 865 * 866 * Sets skb dst, assuming a reference was taken on dst and should 867 * be released by skb_dst_drop() 868 */ 869 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 870 { 871 skb->_skb_refdst = (unsigned long)dst; 872 } 873 874 /** 875 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 876 * @skb: buffer 877 * @dst: dst entry 878 * 879 * Sets skb dst, assuming a reference was not taken on dst. 880 * If dst entry is cached, we do not take reference and dst_release 881 * will be avoided by refdst_drop. If dst entry is not cached, we take 882 * reference, so that last dst_release can destroy the dst immediately. 883 */ 884 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 885 { 886 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 887 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 888 } 889 890 /** 891 * skb_dst_is_noref - Test if skb dst isn't refcounted 892 * @skb: buffer 893 */ 894 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 895 { 896 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 897 } 898 899 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 900 { 901 return (struct rtable *)skb_dst(skb); 902 } 903 904 /* For mangling skb->pkt_type from user space side from applications 905 * such as nft, tc, etc, we only allow a conservative subset of 906 * possible pkt_types to be set. 907 */ 908 static inline bool skb_pkt_type_ok(u32 ptype) 909 { 910 return ptype <= PACKET_OTHERHOST; 911 } 912 913 void kfree_skb(struct sk_buff *skb); 914 void kfree_skb_list(struct sk_buff *segs); 915 void skb_tx_error(struct sk_buff *skb); 916 void consume_skb(struct sk_buff *skb); 917 void __kfree_skb(struct sk_buff *skb); 918 extern struct kmem_cache *skbuff_head_cache; 919 920 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 921 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 922 bool *fragstolen, int *delta_truesize); 923 924 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 925 int node); 926 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 927 struct sk_buff *build_skb(void *data, unsigned int frag_size); 928 static inline struct sk_buff *alloc_skb(unsigned int size, 929 gfp_t priority) 930 { 931 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 932 } 933 934 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 935 unsigned long data_len, 936 int max_page_order, 937 int *errcode, 938 gfp_t gfp_mask); 939 940 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 941 struct sk_buff_fclones { 942 struct sk_buff skb1; 943 944 struct sk_buff skb2; 945 946 atomic_t fclone_ref; 947 }; 948 949 /** 950 * skb_fclone_busy - check if fclone is busy 951 * @sk: socket 952 * @skb: buffer 953 * 954 * Returns true if skb is a fast clone, and its clone is not freed. 955 * Some drivers call skb_orphan() in their ndo_start_xmit(), 956 * so we also check that this didnt happen. 957 */ 958 static inline bool skb_fclone_busy(const struct sock *sk, 959 const struct sk_buff *skb) 960 { 961 const struct sk_buff_fclones *fclones; 962 963 fclones = container_of(skb, struct sk_buff_fclones, skb1); 964 965 return skb->fclone == SKB_FCLONE_ORIG && 966 atomic_read(&fclones->fclone_ref) > 1 && 967 fclones->skb2.sk == sk; 968 } 969 970 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 971 gfp_t priority) 972 { 973 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 974 } 975 976 struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 977 static inline struct sk_buff *alloc_skb_head(gfp_t priority) 978 { 979 return __alloc_skb_head(priority, -1); 980 } 981 982 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 983 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 984 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 985 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 986 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 987 gfp_t gfp_mask, bool fclone); 988 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 989 gfp_t gfp_mask) 990 { 991 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 992 } 993 994 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 995 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 996 unsigned int headroom); 997 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 998 int newtailroom, gfp_t priority); 999 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1000 int offset, int len); 1001 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 1002 int len); 1003 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1004 int skb_pad(struct sk_buff *skb, int pad); 1005 #define dev_kfree_skb(a) consume_skb(a) 1006 1007 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1008 int getfrag(void *from, char *to, int offset, 1009 int len, int odd, struct sk_buff *skb), 1010 void *from, int length); 1011 1012 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1013 int offset, size_t size); 1014 1015 struct skb_seq_state { 1016 __u32 lower_offset; 1017 __u32 upper_offset; 1018 __u32 frag_idx; 1019 __u32 stepped_offset; 1020 struct sk_buff *root_skb; 1021 struct sk_buff *cur_skb; 1022 __u8 *frag_data; 1023 }; 1024 1025 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1026 unsigned int to, struct skb_seq_state *st); 1027 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1028 struct skb_seq_state *st); 1029 void skb_abort_seq_read(struct skb_seq_state *st); 1030 1031 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1032 unsigned int to, struct ts_config *config); 1033 1034 /* 1035 * Packet hash types specify the type of hash in skb_set_hash. 1036 * 1037 * Hash types refer to the protocol layer addresses which are used to 1038 * construct a packet's hash. The hashes are used to differentiate or identify 1039 * flows of the protocol layer for the hash type. Hash types are either 1040 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1041 * 1042 * Properties of hashes: 1043 * 1044 * 1) Two packets in different flows have different hash values 1045 * 2) Two packets in the same flow should have the same hash value 1046 * 1047 * A hash at a higher layer is considered to be more specific. A driver should 1048 * set the most specific hash possible. 1049 * 1050 * A driver cannot indicate a more specific hash than the layer at which a hash 1051 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1052 * 1053 * A driver may indicate a hash level which is less specific than the 1054 * actual layer the hash was computed on. For instance, a hash computed 1055 * at L4 may be considered an L3 hash. This should only be done if the 1056 * driver can't unambiguously determine that the HW computed the hash at 1057 * the higher layer. Note that the "should" in the second property above 1058 * permits this. 1059 */ 1060 enum pkt_hash_types { 1061 PKT_HASH_TYPE_NONE, /* Undefined type */ 1062 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1063 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1064 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1065 }; 1066 1067 static inline void skb_clear_hash(struct sk_buff *skb) 1068 { 1069 skb->hash = 0; 1070 skb->sw_hash = 0; 1071 skb->l4_hash = 0; 1072 } 1073 1074 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1075 { 1076 if (!skb->l4_hash) 1077 skb_clear_hash(skb); 1078 } 1079 1080 static inline void 1081 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1082 { 1083 skb->l4_hash = is_l4; 1084 skb->sw_hash = is_sw; 1085 skb->hash = hash; 1086 } 1087 1088 static inline void 1089 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1090 { 1091 /* Used by drivers to set hash from HW */ 1092 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1093 } 1094 1095 static inline void 1096 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1097 { 1098 __skb_set_hash(skb, hash, true, is_l4); 1099 } 1100 1101 void __skb_get_hash(struct sk_buff *skb); 1102 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1103 u32 skb_get_poff(const struct sk_buff *skb); 1104 u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1105 const struct flow_keys *keys, int hlen); 1106 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1107 void *data, int hlen_proto); 1108 1109 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1110 int thoff, u8 ip_proto) 1111 { 1112 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1113 } 1114 1115 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1116 const struct flow_dissector_key *key, 1117 unsigned int key_count); 1118 1119 bool __skb_flow_dissect(const struct sk_buff *skb, 1120 struct flow_dissector *flow_dissector, 1121 void *target_container, 1122 void *data, __be16 proto, int nhoff, int hlen, 1123 unsigned int flags); 1124 1125 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1126 struct flow_dissector *flow_dissector, 1127 void *target_container, unsigned int flags) 1128 { 1129 return __skb_flow_dissect(skb, flow_dissector, target_container, 1130 NULL, 0, 0, 0, flags); 1131 } 1132 1133 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1134 struct flow_keys *flow, 1135 unsigned int flags) 1136 { 1137 memset(flow, 0, sizeof(*flow)); 1138 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1139 NULL, 0, 0, 0, flags); 1140 } 1141 1142 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1143 void *data, __be16 proto, 1144 int nhoff, int hlen, 1145 unsigned int flags) 1146 { 1147 memset(flow, 0, sizeof(*flow)); 1148 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1149 data, proto, nhoff, hlen, flags); 1150 } 1151 1152 static inline __u32 skb_get_hash(struct sk_buff *skb) 1153 { 1154 if (!skb->l4_hash && !skb->sw_hash) 1155 __skb_get_hash(skb); 1156 1157 return skb->hash; 1158 } 1159 1160 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1161 1162 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1163 { 1164 if (!skb->l4_hash && !skb->sw_hash) { 1165 struct flow_keys keys; 1166 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1167 1168 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1169 } 1170 1171 return skb->hash; 1172 } 1173 1174 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1175 1176 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1177 { 1178 if (!skb->l4_hash && !skb->sw_hash) { 1179 struct flow_keys keys; 1180 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1181 1182 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1183 } 1184 1185 return skb->hash; 1186 } 1187 1188 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1189 1190 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1191 { 1192 return skb->hash; 1193 } 1194 1195 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1196 { 1197 to->hash = from->hash; 1198 to->sw_hash = from->sw_hash; 1199 to->l4_hash = from->l4_hash; 1200 }; 1201 1202 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1203 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1204 { 1205 return skb->head + skb->end; 1206 } 1207 1208 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1209 { 1210 return skb->end; 1211 } 1212 #else 1213 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1214 { 1215 return skb->end; 1216 } 1217 1218 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1219 { 1220 return skb->end - skb->head; 1221 } 1222 #endif 1223 1224 /* Internal */ 1225 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1226 1227 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1228 { 1229 return &skb_shinfo(skb)->hwtstamps; 1230 } 1231 1232 /** 1233 * skb_queue_empty - check if a queue is empty 1234 * @list: queue head 1235 * 1236 * Returns true if the queue is empty, false otherwise. 1237 */ 1238 static inline int skb_queue_empty(const struct sk_buff_head *list) 1239 { 1240 return list->next == (const struct sk_buff *) list; 1241 } 1242 1243 /** 1244 * skb_queue_is_last - check if skb is the last entry in the queue 1245 * @list: queue head 1246 * @skb: buffer 1247 * 1248 * Returns true if @skb is the last buffer on the list. 1249 */ 1250 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1251 const struct sk_buff *skb) 1252 { 1253 return skb->next == (const struct sk_buff *) list; 1254 } 1255 1256 /** 1257 * skb_queue_is_first - check if skb is the first entry in the queue 1258 * @list: queue head 1259 * @skb: buffer 1260 * 1261 * Returns true if @skb is the first buffer on the list. 1262 */ 1263 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1264 const struct sk_buff *skb) 1265 { 1266 return skb->prev == (const struct sk_buff *) list; 1267 } 1268 1269 /** 1270 * skb_queue_next - return the next packet in the queue 1271 * @list: queue head 1272 * @skb: current buffer 1273 * 1274 * Return the next packet in @list after @skb. It is only valid to 1275 * call this if skb_queue_is_last() evaluates to false. 1276 */ 1277 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1278 const struct sk_buff *skb) 1279 { 1280 /* This BUG_ON may seem severe, but if we just return then we 1281 * are going to dereference garbage. 1282 */ 1283 BUG_ON(skb_queue_is_last(list, skb)); 1284 return skb->next; 1285 } 1286 1287 /** 1288 * skb_queue_prev - return the prev packet in the queue 1289 * @list: queue head 1290 * @skb: current buffer 1291 * 1292 * Return the prev packet in @list before @skb. It is only valid to 1293 * call this if skb_queue_is_first() evaluates to false. 1294 */ 1295 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1296 const struct sk_buff *skb) 1297 { 1298 /* This BUG_ON may seem severe, but if we just return then we 1299 * are going to dereference garbage. 1300 */ 1301 BUG_ON(skb_queue_is_first(list, skb)); 1302 return skb->prev; 1303 } 1304 1305 /** 1306 * skb_get - reference buffer 1307 * @skb: buffer to reference 1308 * 1309 * Makes another reference to a socket buffer and returns a pointer 1310 * to the buffer. 1311 */ 1312 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1313 { 1314 atomic_inc(&skb->users); 1315 return skb; 1316 } 1317 1318 /* 1319 * If users == 1, we are the only owner and are can avoid redundant 1320 * atomic change. 1321 */ 1322 1323 /** 1324 * skb_cloned - is the buffer a clone 1325 * @skb: buffer to check 1326 * 1327 * Returns true if the buffer was generated with skb_clone() and is 1328 * one of multiple shared copies of the buffer. Cloned buffers are 1329 * shared data so must not be written to under normal circumstances. 1330 */ 1331 static inline int skb_cloned(const struct sk_buff *skb) 1332 { 1333 return skb->cloned && 1334 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1335 } 1336 1337 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1338 { 1339 might_sleep_if(gfpflags_allow_blocking(pri)); 1340 1341 if (skb_cloned(skb)) 1342 return pskb_expand_head(skb, 0, 0, pri); 1343 1344 return 0; 1345 } 1346 1347 /** 1348 * skb_header_cloned - is the header a clone 1349 * @skb: buffer to check 1350 * 1351 * Returns true if modifying the header part of the buffer requires 1352 * the data to be copied. 1353 */ 1354 static inline int skb_header_cloned(const struct sk_buff *skb) 1355 { 1356 int dataref; 1357 1358 if (!skb->cloned) 1359 return 0; 1360 1361 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1362 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1363 return dataref != 1; 1364 } 1365 1366 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1367 { 1368 might_sleep_if(gfpflags_allow_blocking(pri)); 1369 1370 if (skb_header_cloned(skb)) 1371 return pskb_expand_head(skb, 0, 0, pri); 1372 1373 return 0; 1374 } 1375 1376 /** 1377 * skb_header_release - release reference to header 1378 * @skb: buffer to operate on 1379 * 1380 * Drop a reference to the header part of the buffer. This is done 1381 * by acquiring a payload reference. You must not read from the header 1382 * part of skb->data after this. 1383 * Note : Check if you can use __skb_header_release() instead. 1384 */ 1385 static inline void skb_header_release(struct sk_buff *skb) 1386 { 1387 BUG_ON(skb->nohdr); 1388 skb->nohdr = 1; 1389 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1390 } 1391 1392 /** 1393 * __skb_header_release - release reference to header 1394 * @skb: buffer to operate on 1395 * 1396 * Variant of skb_header_release() assuming skb is private to caller. 1397 * We can avoid one atomic operation. 1398 */ 1399 static inline void __skb_header_release(struct sk_buff *skb) 1400 { 1401 skb->nohdr = 1; 1402 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1403 } 1404 1405 1406 /** 1407 * skb_shared - is the buffer shared 1408 * @skb: buffer to check 1409 * 1410 * Returns true if more than one person has a reference to this 1411 * buffer. 1412 */ 1413 static inline int skb_shared(const struct sk_buff *skb) 1414 { 1415 return atomic_read(&skb->users) != 1; 1416 } 1417 1418 /** 1419 * skb_share_check - check if buffer is shared and if so clone it 1420 * @skb: buffer to check 1421 * @pri: priority for memory allocation 1422 * 1423 * If the buffer is shared the buffer is cloned and the old copy 1424 * drops a reference. A new clone with a single reference is returned. 1425 * If the buffer is not shared the original buffer is returned. When 1426 * being called from interrupt status or with spinlocks held pri must 1427 * be GFP_ATOMIC. 1428 * 1429 * NULL is returned on a memory allocation failure. 1430 */ 1431 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1432 { 1433 might_sleep_if(gfpflags_allow_blocking(pri)); 1434 if (skb_shared(skb)) { 1435 struct sk_buff *nskb = skb_clone(skb, pri); 1436 1437 if (likely(nskb)) 1438 consume_skb(skb); 1439 else 1440 kfree_skb(skb); 1441 skb = nskb; 1442 } 1443 return skb; 1444 } 1445 1446 /* 1447 * Copy shared buffers into a new sk_buff. We effectively do COW on 1448 * packets to handle cases where we have a local reader and forward 1449 * and a couple of other messy ones. The normal one is tcpdumping 1450 * a packet thats being forwarded. 1451 */ 1452 1453 /** 1454 * skb_unshare - make a copy of a shared buffer 1455 * @skb: buffer to check 1456 * @pri: priority for memory allocation 1457 * 1458 * If the socket buffer is a clone then this function creates a new 1459 * copy of the data, drops a reference count on the old copy and returns 1460 * the new copy with the reference count at 1. If the buffer is not a clone 1461 * the original buffer is returned. When called with a spinlock held or 1462 * from interrupt state @pri must be %GFP_ATOMIC 1463 * 1464 * %NULL is returned on a memory allocation failure. 1465 */ 1466 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1467 gfp_t pri) 1468 { 1469 might_sleep_if(gfpflags_allow_blocking(pri)); 1470 if (skb_cloned(skb)) { 1471 struct sk_buff *nskb = skb_copy(skb, pri); 1472 1473 /* Free our shared copy */ 1474 if (likely(nskb)) 1475 consume_skb(skb); 1476 else 1477 kfree_skb(skb); 1478 skb = nskb; 1479 } 1480 return skb; 1481 } 1482 1483 /** 1484 * skb_peek - peek at the head of an &sk_buff_head 1485 * @list_: list to peek at 1486 * 1487 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1488 * be careful with this one. A peek leaves the buffer on the 1489 * list and someone else may run off with it. You must hold 1490 * the appropriate locks or have a private queue to do this. 1491 * 1492 * Returns %NULL for an empty list or a pointer to the head element. 1493 * The reference count is not incremented and the reference is therefore 1494 * volatile. Use with caution. 1495 */ 1496 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1497 { 1498 struct sk_buff *skb = list_->next; 1499 1500 if (skb == (struct sk_buff *)list_) 1501 skb = NULL; 1502 return skb; 1503 } 1504 1505 /** 1506 * skb_peek_next - peek skb following the given one from a queue 1507 * @skb: skb to start from 1508 * @list_: list to peek at 1509 * 1510 * Returns %NULL when the end of the list is met or a pointer to the 1511 * next element. The reference count is not incremented and the 1512 * reference is therefore volatile. Use with caution. 1513 */ 1514 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1515 const struct sk_buff_head *list_) 1516 { 1517 struct sk_buff *next = skb->next; 1518 1519 if (next == (struct sk_buff *)list_) 1520 next = NULL; 1521 return next; 1522 } 1523 1524 /** 1525 * skb_peek_tail - peek at the tail of an &sk_buff_head 1526 * @list_: list to peek at 1527 * 1528 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1529 * be careful with this one. A peek leaves the buffer on the 1530 * list and someone else may run off with it. You must hold 1531 * the appropriate locks or have a private queue to do this. 1532 * 1533 * Returns %NULL for an empty list or a pointer to the tail element. 1534 * The reference count is not incremented and the reference is therefore 1535 * volatile. Use with caution. 1536 */ 1537 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1538 { 1539 struct sk_buff *skb = list_->prev; 1540 1541 if (skb == (struct sk_buff *)list_) 1542 skb = NULL; 1543 return skb; 1544 1545 } 1546 1547 /** 1548 * skb_queue_len - get queue length 1549 * @list_: list to measure 1550 * 1551 * Return the length of an &sk_buff queue. 1552 */ 1553 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1554 { 1555 return list_->qlen; 1556 } 1557 1558 /** 1559 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1560 * @list: queue to initialize 1561 * 1562 * This initializes only the list and queue length aspects of 1563 * an sk_buff_head object. This allows to initialize the list 1564 * aspects of an sk_buff_head without reinitializing things like 1565 * the spinlock. It can also be used for on-stack sk_buff_head 1566 * objects where the spinlock is known to not be used. 1567 */ 1568 static inline void __skb_queue_head_init(struct sk_buff_head *list) 1569 { 1570 list->prev = list->next = (struct sk_buff *)list; 1571 list->qlen = 0; 1572 } 1573 1574 /* 1575 * This function creates a split out lock class for each invocation; 1576 * this is needed for now since a whole lot of users of the skb-queue 1577 * infrastructure in drivers have different locking usage (in hardirq) 1578 * than the networking core (in softirq only). In the long run either the 1579 * network layer or drivers should need annotation to consolidate the 1580 * main types of usage into 3 classes. 1581 */ 1582 static inline void skb_queue_head_init(struct sk_buff_head *list) 1583 { 1584 spin_lock_init(&list->lock); 1585 __skb_queue_head_init(list); 1586 } 1587 1588 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1589 struct lock_class_key *class) 1590 { 1591 skb_queue_head_init(list); 1592 lockdep_set_class(&list->lock, class); 1593 } 1594 1595 /* 1596 * Insert an sk_buff on a list. 1597 * 1598 * The "__skb_xxxx()" functions are the non-atomic ones that 1599 * can only be called with interrupts disabled. 1600 */ 1601 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1602 struct sk_buff_head *list); 1603 static inline void __skb_insert(struct sk_buff *newsk, 1604 struct sk_buff *prev, struct sk_buff *next, 1605 struct sk_buff_head *list) 1606 { 1607 newsk->next = next; 1608 newsk->prev = prev; 1609 next->prev = prev->next = newsk; 1610 list->qlen++; 1611 } 1612 1613 static inline void __skb_queue_splice(const struct sk_buff_head *list, 1614 struct sk_buff *prev, 1615 struct sk_buff *next) 1616 { 1617 struct sk_buff *first = list->next; 1618 struct sk_buff *last = list->prev; 1619 1620 first->prev = prev; 1621 prev->next = first; 1622 1623 last->next = next; 1624 next->prev = last; 1625 } 1626 1627 /** 1628 * skb_queue_splice - join two skb lists, this is designed for stacks 1629 * @list: the new list to add 1630 * @head: the place to add it in the first list 1631 */ 1632 static inline void skb_queue_splice(const struct sk_buff_head *list, 1633 struct sk_buff_head *head) 1634 { 1635 if (!skb_queue_empty(list)) { 1636 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1637 head->qlen += list->qlen; 1638 } 1639 } 1640 1641 /** 1642 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1643 * @list: the new list to add 1644 * @head: the place to add it in the first list 1645 * 1646 * The list at @list is reinitialised 1647 */ 1648 static inline void skb_queue_splice_init(struct sk_buff_head *list, 1649 struct sk_buff_head *head) 1650 { 1651 if (!skb_queue_empty(list)) { 1652 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1653 head->qlen += list->qlen; 1654 __skb_queue_head_init(list); 1655 } 1656 } 1657 1658 /** 1659 * skb_queue_splice_tail - join two skb lists, each list being a queue 1660 * @list: the new list to add 1661 * @head: the place to add it in the first list 1662 */ 1663 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1664 struct sk_buff_head *head) 1665 { 1666 if (!skb_queue_empty(list)) { 1667 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1668 head->qlen += list->qlen; 1669 } 1670 } 1671 1672 /** 1673 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1674 * @list: the new list to add 1675 * @head: the place to add it in the first list 1676 * 1677 * Each of the lists is a queue. 1678 * The list at @list is reinitialised 1679 */ 1680 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1681 struct sk_buff_head *head) 1682 { 1683 if (!skb_queue_empty(list)) { 1684 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1685 head->qlen += list->qlen; 1686 __skb_queue_head_init(list); 1687 } 1688 } 1689 1690 /** 1691 * __skb_queue_after - queue a buffer at the list head 1692 * @list: list to use 1693 * @prev: place after this buffer 1694 * @newsk: buffer to queue 1695 * 1696 * Queue a buffer int the middle of a list. This function takes no locks 1697 * and you must therefore hold required locks before calling it. 1698 * 1699 * A buffer cannot be placed on two lists at the same time. 1700 */ 1701 static inline void __skb_queue_after(struct sk_buff_head *list, 1702 struct sk_buff *prev, 1703 struct sk_buff *newsk) 1704 { 1705 __skb_insert(newsk, prev, prev->next, list); 1706 } 1707 1708 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1709 struct sk_buff_head *list); 1710 1711 static inline void __skb_queue_before(struct sk_buff_head *list, 1712 struct sk_buff *next, 1713 struct sk_buff *newsk) 1714 { 1715 __skb_insert(newsk, next->prev, next, list); 1716 } 1717 1718 /** 1719 * __skb_queue_head - queue a buffer at the list head 1720 * @list: list to use 1721 * @newsk: buffer to queue 1722 * 1723 * Queue a buffer at the start of a list. This function takes no locks 1724 * and you must therefore hold required locks before calling it. 1725 * 1726 * A buffer cannot be placed on two lists at the same time. 1727 */ 1728 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1729 static inline void __skb_queue_head(struct sk_buff_head *list, 1730 struct sk_buff *newsk) 1731 { 1732 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1733 } 1734 1735 /** 1736 * __skb_queue_tail - queue a buffer at the list tail 1737 * @list: list to use 1738 * @newsk: buffer to queue 1739 * 1740 * Queue a buffer at the end of a list. This function takes no locks 1741 * and you must therefore hold required locks before calling it. 1742 * 1743 * A buffer cannot be placed on two lists at the same time. 1744 */ 1745 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1746 static inline void __skb_queue_tail(struct sk_buff_head *list, 1747 struct sk_buff *newsk) 1748 { 1749 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1750 } 1751 1752 /* 1753 * remove sk_buff from list. _Must_ be called atomically, and with 1754 * the list known.. 1755 */ 1756 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1757 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1758 { 1759 struct sk_buff *next, *prev; 1760 1761 list->qlen--; 1762 next = skb->next; 1763 prev = skb->prev; 1764 skb->next = skb->prev = NULL; 1765 next->prev = prev; 1766 prev->next = next; 1767 } 1768 1769 /** 1770 * __skb_dequeue - remove from the head of the queue 1771 * @list: list to dequeue from 1772 * 1773 * Remove the head of the list. This function does not take any locks 1774 * so must be used with appropriate locks held only. The head item is 1775 * returned or %NULL if the list is empty. 1776 */ 1777 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1778 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1779 { 1780 struct sk_buff *skb = skb_peek(list); 1781 if (skb) 1782 __skb_unlink(skb, list); 1783 return skb; 1784 } 1785 1786 /** 1787 * __skb_dequeue_tail - remove from the tail of the queue 1788 * @list: list to dequeue from 1789 * 1790 * Remove the tail of the list. This function does not take any locks 1791 * so must be used with appropriate locks held only. The tail item is 1792 * returned or %NULL if the list is empty. 1793 */ 1794 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1795 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1796 { 1797 struct sk_buff *skb = skb_peek_tail(list); 1798 if (skb) 1799 __skb_unlink(skb, list); 1800 return skb; 1801 } 1802 1803 1804 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1805 { 1806 return skb->data_len; 1807 } 1808 1809 static inline unsigned int skb_headlen(const struct sk_buff *skb) 1810 { 1811 return skb->len - skb->data_len; 1812 } 1813 1814 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1815 { 1816 unsigned int i, len = 0; 1817 1818 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1819 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1820 return len + skb_headlen(skb); 1821 } 1822 1823 /** 1824 * __skb_fill_page_desc - initialise a paged fragment in an skb 1825 * @skb: buffer containing fragment to be initialised 1826 * @i: paged fragment index to initialise 1827 * @page: the page to use for this fragment 1828 * @off: the offset to the data with @page 1829 * @size: the length of the data 1830 * 1831 * Initialises the @i'th fragment of @skb to point to &size bytes at 1832 * offset @off within @page. 1833 * 1834 * Does not take any additional reference on the fragment. 1835 */ 1836 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1837 struct page *page, int off, int size) 1838 { 1839 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1840 1841 /* 1842 * Propagate page pfmemalloc to the skb if we can. The problem is 1843 * that not all callers have unique ownership of the page but rely 1844 * on page_is_pfmemalloc doing the right thing(tm). 1845 */ 1846 frag->page.p = page; 1847 frag->page_offset = off; 1848 skb_frag_size_set(frag, size); 1849 1850 page = compound_head(page); 1851 if (page_is_pfmemalloc(page)) 1852 skb->pfmemalloc = true; 1853 } 1854 1855 /** 1856 * skb_fill_page_desc - initialise a paged fragment in an skb 1857 * @skb: buffer containing fragment to be initialised 1858 * @i: paged fragment index to initialise 1859 * @page: the page to use for this fragment 1860 * @off: the offset to the data with @page 1861 * @size: the length of the data 1862 * 1863 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1864 * @skb to point to @size bytes at offset @off within @page. In 1865 * addition updates @skb such that @i is the last fragment. 1866 * 1867 * Does not take any additional reference on the fragment. 1868 */ 1869 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1870 struct page *page, int off, int size) 1871 { 1872 __skb_fill_page_desc(skb, i, page, off, size); 1873 skb_shinfo(skb)->nr_frags = i + 1; 1874 } 1875 1876 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1877 int size, unsigned int truesize); 1878 1879 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1880 unsigned int truesize); 1881 1882 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1883 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1884 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1885 1886 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1887 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1888 { 1889 return skb->head + skb->tail; 1890 } 1891 1892 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1893 { 1894 skb->tail = skb->data - skb->head; 1895 } 1896 1897 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1898 { 1899 skb_reset_tail_pointer(skb); 1900 skb->tail += offset; 1901 } 1902 1903 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 1904 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1905 { 1906 return skb->tail; 1907 } 1908 1909 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1910 { 1911 skb->tail = skb->data; 1912 } 1913 1914 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1915 { 1916 skb->tail = skb->data + offset; 1917 } 1918 1919 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1920 1921 /* 1922 * Add data to an sk_buff 1923 */ 1924 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1925 unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1926 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1927 { 1928 unsigned char *tmp = skb_tail_pointer(skb); 1929 SKB_LINEAR_ASSERT(skb); 1930 skb->tail += len; 1931 skb->len += len; 1932 return tmp; 1933 } 1934 1935 unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1936 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1937 { 1938 skb->data -= len; 1939 skb->len += len; 1940 return skb->data; 1941 } 1942 1943 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1944 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1945 { 1946 skb->len -= len; 1947 BUG_ON(skb->len < skb->data_len); 1948 return skb->data += len; 1949 } 1950 1951 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1952 { 1953 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1954 } 1955 1956 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1957 1958 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1959 { 1960 if (len > skb_headlen(skb) && 1961 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1962 return NULL; 1963 skb->len -= len; 1964 return skb->data += len; 1965 } 1966 1967 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1968 { 1969 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1970 } 1971 1972 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1973 { 1974 if (likely(len <= skb_headlen(skb))) 1975 return 1; 1976 if (unlikely(len > skb->len)) 1977 return 0; 1978 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1979 } 1980 1981 void skb_condense(struct sk_buff *skb); 1982 1983 /** 1984 * skb_headroom - bytes at buffer head 1985 * @skb: buffer to check 1986 * 1987 * Return the number of bytes of free space at the head of an &sk_buff. 1988 */ 1989 static inline unsigned int skb_headroom(const struct sk_buff *skb) 1990 { 1991 return skb->data - skb->head; 1992 } 1993 1994 /** 1995 * skb_tailroom - bytes at buffer end 1996 * @skb: buffer to check 1997 * 1998 * Return the number of bytes of free space at the tail of an sk_buff 1999 */ 2000 static inline int skb_tailroom(const struct sk_buff *skb) 2001 { 2002 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2003 } 2004 2005 /** 2006 * skb_availroom - bytes at buffer end 2007 * @skb: buffer to check 2008 * 2009 * Return the number of bytes of free space at the tail of an sk_buff 2010 * allocated by sk_stream_alloc() 2011 */ 2012 static inline int skb_availroom(const struct sk_buff *skb) 2013 { 2014 if (skb_is_nonlinear(skb)) 2015 return 0; 2016 2017 return skb->end - skb->tail - skb->reserved_tailroom; 2018 } 2019 2020 /** 2021 * skb_reserve - adjust headroom 2022 * @skb: buffer to alter 2023 * @len: bytes to move 2024 * 2025 * Increase the headroom of an empty &sk_buff by reducing the tail 2026 * room. This is only allowed for an empty buffer. 2027 */ 2028 static inline void skb_reserve(struct sk_buff *skb, int len) 2029 { 2030 skb->data += len; 2031 skb->tail += len; 2032 } 2033 2034 /** 2035 * skb_tailroom_reserve - adjust reserved_tailroom 2036 * @skb: buffer to alter 2037 * @mtu: maximum amount of headlen permitted 2038 * @needed_tailroom: minimum amount of reserved_tailroom 2039 * 2040 * Set reserved_tailroom so that headlen can be as large as possible but 2041 * not larger than mtu and tailroom cannot be smaller than 2042 * needed_tailroom. 2043 * The required headroom should already have been reserved before using 2044 * this function. 2045 */ 2046 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2047 unsigned int needed_tailroom) 2048 { 2049 SKB_LINEAR_ASSERT(skb); 2050 if (mtu < skb_tailroom(skb) - needed_tailroom) 2051 /* use at most mtu */ 2052 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2053 else 2054 /* use up to all available space */ 2055 skb->reserved_tailroom = needed_tailroom; 2056 } 2057 2058 #define ENCAP_TYPE_ETHER 0 2059 #define ENCAP_TYPE_IPPROTO 1 2060 2061 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2062 __be16 protocol) 2063 { 2064 skb->inner_protocol = protocol; 2065 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2066 } 2067 2068 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2069 __u8 ipproto) 2070 { 2071 skb->inner_ipproto = ipproto; 2072 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2073 } 2074 2075 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2076 { 2077 skb->inner_mac_header = skb->mac_header; 2078 skb->inner_network_header = skb->network_header; 2079 skb->inner_transport_header = skb->transport_header; 2080 } 2081 2082 static inline void skb_reset_mac_len(struct sk_buff *skb) 2083 { 2084 skb->mac_len = skb->network_header - skb->mac_header; 2085 } 2086 2087 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2088 *skb) 2089 { 2090 return skb->head + skb->inner_transport_header; 2091 } 2092 2093 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2094 { 2095 return skb_inner_transport_header(skb) - skb->data; 2096 } 2097 2098 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2099 { 2100 skb->inner_transport_header = skb->data - skb->head; 2101 } 2102 2103 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2104 const int offset) 2105 { 2106 skb_reset_inner_transport_header(skb); 2107 skb->inner_transport_header += offset; 2108 } 2109 2110 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2111 { 2112 return skb->head + skb->inner_network_header; 2113 } 2114 2115 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2116 { 2117 skb->inner_network_header = skb->data - skb->head; 2118 } 2119 2120 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2121 const int offset) 2122 { 2123 skb_reset_inner_network_header(skb); 2124 skb->inner_network_header += offset; 2125 } 2126 2127 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2128 { 2129 return skb->head + skb->inner_mac_header; 2130 } 2131 2132 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2133 { 2134 skb->inner_mac_header = skb->data - skb->head; 2135 } 2136 2137 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2138 const int offset) 2139 { 2140 skb_reset_inner_mac_header(skb); 2141 skb->inner_mac_header += offset; 2142 } 2143 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2144 { 2145 return skb->transport_header != (typeof(skb->transport_header))~0U; 2146 } 2147 2148 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2149 { 2150 return skb->head + skb->transport_header; 2151 } 2152 2153 static inline void skb_reset_transport_header(struct sk_buff *skb) 2154 { 2155 skb->transport_header = skb->data - skb->head; 2156 } 2157 2158 static inline void skb_set_transport_header(struct sk_buff *skb, 2159 const int offset) 2160 { 2161 skb_reset_transport_header(skb); 2162 skb->transport_header += offset; 2163 } 2164 2165 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2166 { 2167 return skb->head + skb->network_header; 2168 } 2169 2170 static inline void skb_reset_network_header(struct sk_buff *skb) 2171 { 2172 skb->network_header = skb->data - skb->head; 2173 } 2174 2175 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2176 { 2177 skb_reset_network_header(skb); 2178 skb->network_header += offset; 2179 } 2180 2181 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2182 { 2183 return skb->head + skb->mac_header; 2184 } 2185 2186 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2187 { 2188 return skb->mac_header != (typeof(skb->mac_header))~0U; 2189 } 2190 2191 static inline void skb_reset_mac_header(struct sk_buff *skb) 2192 { 2193 skb->mac_header = skb->data - skb->head; 2194 } 2195 2196 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2197 { 2198 skb_reset_mac_header(skb); 2199 skb->mac_header += offset; 2200 } 2201 2202 static inline void skb_pop_mac_header(struct sk_buff *skb) 2203 { 2204 skb->mac_header = skb->network_header; 2205 } 2206 2207 static inline void skb_probe_transport_header(struct sk_buff *skb, 2208 const int offset_hint) 2209 { 2210 struct flow_keys keys; 2211 2212 if (skb_transport_header_was_set(skb)) 2213 return; 2214 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2215 skb_set_transport_header(skb, keys.control.thoff); 2216 else 2217 skb_set_transport_header(skb, offset_hint); 2218 } 2219 2220 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2221 { 2222 if (skb_mac_header_was_set(skb)) { 2223 const unsigned char *old_mac = skb_mac_header(skb); 2224 2225 skb_set_mac_header(skb, -skb->mac_len); 2226 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2227 } 2228 } 2229 2230 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2231 { 2232 return skb->csum_start - skb_headroom(skb); 2233 } 2234 2235 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2236 { 2237 return skb->head + skb->csum_start; 2238 } 2239 2240 static inline int skb_transport_offset(const struct sk_buff *skb) 2241 { 2242 return skb_transport_header(skb) - skb->data; 2243 } 2244 2245 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2246 { 2247 return skb->transport_header - skb->network_header; 2248 } 2249 2250 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2251 { 2252 return skb->inner_transport_header - skb->inner_network_header; 2253 } 2254 2255 static inline int skb_network_offset(const struct sk_buff *skb) 2256 { 2257 return skb_network_header(skb) - skb->data; 2258 } 2259 2260 static inline int skb_inner_network_offset(const struct sk_buff *skb) 2261 { 2262 return skb_inner_network_header(skb) - skb->data; 2263 } 2264 2265 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2266 { 2267 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2268 } 2269 2270 /* 2271 * CPUs often take a performance hit when accessing unaligned memory 2272 * locations. The actual performance hit varies, it can be small if the 2273 * hardware handles it or large if we have to take an exception and fix it 2274 * in software. 2275 * 2276 * Since an ethernet header is 14 bytes network drivers often end up with 2277 * the IP header at an unaligned offset. The IP header can be aligned by 2278 * shifting the start of the packet by 2 bytes. Drivers should do this 2279 * with: 2280 * 2281 * skb_reserve(skb, NET_IP_ALIGN); 2282 * 2283 * The downside to this alignment of the IP header is that the DMA is now 2284 * unaligned. On some architectures the cost of an unaligned DMA is high 2285 * and this cost outweighs the gains made by aligning the IP header. 2286 * 2287 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2288 * to be overridden. 2289 */ 2290 #ifndef NET_IP_ALIGN 2291 #define NET_IP_ALIGN 2 2292 #endif 2293 2294 /* 2295 * The networking layer reserves some headroom in skb data (via 2296 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2297 * the header has to grow. In the default case, if the header has to grow 2298 * 32 bytes or less we avoid the reallocation. 2299 * 2300 * Unfortunately this headroom changes the DMA alignment of the resulting 2301 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2302 * on some architectures. An architecture can override this value, 2303 * perhaps setting it to a cacheline in size (since that will maintain 2304 * cacheline alignment of the DMA). It must be a power of 2. 2305 * 2306 * Various parts of the networking layer expect at least 32 bytes of 2307 * headroom, you should not reduce this. 2308 * 2309 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2310 * to reduce average number of cache lines per packet. 2311 * get_rps_cpus() for example only access one 64 bytes aligned block : 2312 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2313 */ 2314 #ifndef NET_SKB_PAD 2315 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2316 #endif 2317 2318 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2319 2320 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2321 { 2322 if (unlikely(skb_is_nonlinear(skb))) { 2323 WARN_ON(1); 2324 return; 2325 } 2326 skb->len = len; 2327 skb_set_tail_pointer(skb, len); 2328 } 2329 2330 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2331 { 2332 __skb_set_length(skb, len); 2333 } 2334 2335 void skb_trim(struct sk_buff *skb, unsigned int len); 2336 2337 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2338 { 2339 if (skb->data_len) 2340 return ___pskb_trim(skb, len); 2341 __skb_trim(skb, len); 2342 return 0; 2343 } 2344 2345 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2346 { 2347 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2348 } 2349 2350 /** 2351 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2352 * @skb: buffer to alter 2353 * @len: new length 2354 * 2355 * This is identical to pskb_trim except that the caller knows that 2356 * the skb is not cloned so we should never get an error due to out- 2357 * of-memory. 2358 */ 2359 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2360 { 2361 int err = pskb_trim(skb, len); 2362 BUG_ON(err); 2363 } 2364 2365 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2366 { 2367 unsigned int diff = len - skb->len; 2368 2369 if (skb_tailroom(skb) < diff) { 2370 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2371 GFP_ATOMIC); 2372 if (ret) 2373 return ret; 2374 } 2375 __skb_set_length(skb, len); 2376 return 0; 2377 } 2378 2379 /** 2380 * skb_orphan - orphan a buffer 2381 * @skb: buffer to orphan 2382 * 2383 * If a buffer currently has an owner then we call the owner's 2384 * destructor function and make the @skb unowned. The buffer continues 2385 * to exist but is no longer charged to its former owner. 2386 */ 2387 static inline void skb_orphan(struct sk_buff *skb) 2388 { 2389 if (skb->destructor) { 2390 skb->destructor(skb); 2391 skb->destructor = NULL; 2392 skb->sk = NULL; 2393 } else { 2394 BUG_ON(skb->sk); 2395 } 2396 } 2397 2398 /** 2399 * skb_orphan_frags - orphan the frags contained in a buffer 2400 * @skb: buffer to orphan frags from 2401 * @gfp_mask: allocation mask for replacement pages 2402 * 2403 * For each frag in the SKB which needs a destructor (i.e. has an 2404 * owner) create a copy of that frag and release the original 2405 * page by calling the destructor. 2406 */ 2407 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2408 { 2409 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2410 return 0; 2411 return skb_copy_ubufs(skb, gfp_mask); 2412 } 2413 2414 /** 2415 * __skb_queue_purge - empty a list 2416 * @list: list to empty 2417 * 2418 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2419 * the list and one reference dropped. This function does not take the 2420 * list lock and the caller must hold the relevant locks to use it. 2421 */ 2422 void skb_queue_purge(struct sk_buff_head *list); 2423 static inline void __skb_queue_purge(struct sk_buff_head *list) 2424 { 2425 struct sk_buff *skb; 2426 while ((skb = __skb_dequeue(list)) != NULL) 2427 kfree_skb(skb); 2428 } 2429 2430 void skb_rbtree_purge(struct rb_root *root); 2431 2432 void *netdev_alloc_frag(unsigned int fragsz); 2433 2434 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2435 gfp_t gfp_mask); 2436 2437 /** 2438 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2439 * @dev: network device to receive on 2440 * @length: length to allocate 2441 * 2442 * Allocate a new &sk_buff and assign it a usage count of one. The 2443 * buffer has unspecified headroom built in. Users should allocate 2444 * the headroom they think they need without accounting for the 2445 * built in space. The built in space is used for optimisations. 2446 * 2447 * %NULL is returned if there is no free memory. Although this function 2448 * allocates memory it can be called from an interrupt. 2449 */ 2450 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2451 unsigned int length) 2452 { 2453 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2454 } 2455 2456 /* legacy helper around __netdev_alloc_skb() */ 2457 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2458 gfp_t gfp_mask) 2459 { 2460 return __netdev_alloc_skb(NULL, length, gfp_mask); 2461 } 2462 2463 /* legacy helper around netdev_alloc_skb() */ 2464 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2465 { 2466 return netdev_alloc_skb(NULL, length); 2467 } 2468 2469 2470 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2471 unsigned int length, gfp_t gfp) 2472 { 2473 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2474 2475 if (NET_IP_ALIGN && skb) 2476 skb_reserve(skb, NET_IP_ALIGN); 2477 return skb; 2478 } 2479 2480 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2481 unsigned int length) 2482 { 2483 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2484 } 2485 2486 static inline void skb_free_frag(void *addr) 2487 { 2488 page_frag_free(addr); 2489 } 2490 2491 void *napi_alloc_frag(unsigned int fragsz); 2492 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2493 unsigned int length, gfp_t gfp_mask); 2494 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2495 unsigned int length) 2496 { 2497 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2498 } 2499 void napi_consume_skb(struct sk_buff *skb, int budget); 2500 2501 void __kfree_skb_flush(void); 2502 void __kfree_skb_defer(struct sk_buff *skb); 2503 2504 /** 2505 * __dev_alloc_pages - allocate page for network Rx 2506 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2507 * @order: size of the allocation 2508 * 2509 * Allocate a new page. 2510 * 2511 * %NULL is returned if there is no free memory. 2512 */ 2513 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2514 unsigned int order) 2515 { 2516 /* This piece of code contains several assumptions. 2517 * 1. This is for device Rx, therefor a cold page is preferred. 2518 * 2. The expectation is the user wants a compound page. 2519 * 3. If requesting a order 0 page it will not be compound 2520 * due to the check to see if order has a value in prep_new_page 2521 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2522 * code in gfp_to_alloc_flags that should be enforcing this. 2523 */ 2524 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2525 2526 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2527 } 2528 2529 static inline struct page *dev_alloc_pages(unsigned int order) 2530 { 2531 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2532 } 2533 2534 /** 2535 * __dev_alloc_page - allocate a page for network Rx 2536 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2537 * 2538 * Allocate a new page. 2539 * 2540 * %NULL is returned if there is no free memory. 2541 */ 2542 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2543 { 2544 return __dev_alloc_pages(gfp_mask, 0); 2545 } 2546 2547 static inline struct page *dev_alloc_page(void) 2548 { 2549 return dev_alloc_pages(0); 2550 } 2551 2552 /** 2553 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2554 * @page: The page that was allocated from skb_alloc_page 2555 * @skb: The skb that may need pfmemalloc set 2556 */ 2557 static inline void skb_propagate_pfmemalloc(struct page *page, 2558 struct sk_buff *skb) 2559 { 2560 if (page_is_pfmemalloc(page)) 2561 skb->pfmemalloc = true; 2562 } 2563 2564 /** 2565 * skb_frag_page - retrieve the page referred to by a paged fragment 2566 * @frag: the paged fragment 2567 * 2568 * Returns the &struct page associated with @frag. 2569 */ 2570 static inline struct page *skb_frag_page(const skb_frag_t *frag) 2571 { 2572 return frag->page.p; 2573 } 2574 2575 /** 2576 * __skb_frag_ref - take an addition reference on a paged fragment. 2577 * @frag: the paged fragment 2578 * 2579 * Takes an additional reference on the paged fragment @frag. 2580 */ 2581 static inline void __skb_frag_ref(skb_frag_t *frag) 2582 { 2583 get_page(skb_frag_page(frag)); 2584 } 2585 2586 /** 2587 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2588 * @skb: the buffer 2589 * @f: the fragment offset. 2590 * 2591 * Takes an additional reference on the @f'th paged fragment of @skb. 2592 */ 2593 static inline void skb_frag_ref(struct sk_buff *skb, int f) 2594 { 2595 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2596 } 2597 2598 /** 2599 * __skb_frag_unref - release a reference on a paged fragment. 2600 * @frag: the paged fragment 2601 * 2602 * Releases a reference on the paged fragment @frag. 2603 */ 2604 static inline void __skb_frag_unref(skb_frag_t *frag) 2605 { 2606 put_page(skb_frag_page(frag)); 2607 } 2608 2609 /** 2610 * skb_frag_unref - release a reference on a paged fragment of an skb. 2611 * @skb: the buffer 2612 * @f: the fragment offset 2613 * 2614 * Releases a reference on the @f'th paged fragment of @skb. 2615 */ 2616 static inline void skb_frag_unref(struct sk_buff *skb, int f) 2617 { 2618 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2619 } 2620 2621 /** 2622 * skb_frag_address - gets the address of the data contained in a paged fragment 2623 * @frag: the paged fragment buffer 2624 * 2625 * Returns the address of the data within @frag. The page must already 2626 * be mapped. 2627 */ 2628 static inline void *skb_frag_address(const skb_frag_t *frag) 2629 { 2630 return page_address(skb_frag_page(frag)) + frag->page_offset; 2631 } 2632 2633 /** 2634 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2635 * @frag: the paged fragment buffer 2636 * 2637 * Returns the address of the data within @frag. Checks that the page 2638 * is mapped and returns %NULL otherwise. 2639 */ 2640 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2641 { 2642 void *ptr = page_address(skb_frag_page(frag)); 2643 if (unlikely(!ptr)) 2644 return NULL; 2645 2646 return ptr + frag->page_offset; 2647 } 2648 2649 /** 2650 * __skb_frag_set_page - sets the page contained in a paged fragment 2651 * @frag: the paged fragment 2652 * @page: the page to set 2653 * 2654 * Sets the fragment @frag to contain @page. 2655 */ 2656 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2657 { 2658 frag->page.p = page; 2659 } 2660 2661 /** 2662 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2663 * @skb: the buffer 2664 * @f: the fragment offset 2665 * @page: the page to set 2666 * 2667 * Sets the @f'th fragment of @skb to contain @page. 2668 */ 2669 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2670 struct page *page) 2671 { 2672 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2673 } 2674 2675 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2676 2677 /** 2678 * skb_frag_dma_map - maps a paged fragment via the DMA API 2679 * @dev: the device to map the fragment to 2680 * @frag: the paged fragment to map 2681 * @offset: the offset within the fragment (starting at the 2682 * fragment's own offset) 2683 * @size: the number of bytes to map 2684 * @dir: the direction of the mapping (%PCI_DMA_*) 2685 * 2686 * Maps the page associated with @frag to @device. 2687 */ 2688 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2689 const skb_frag_t *frag, 2690 size_t offset, size_t size, 2691 enum dma_data_direction dir) 2692 { 2693 return dma_map_page(dev, skb_frag_page(frag), 2694 frag->page_offset + offset, size, dir); 2695 } 2696 2697 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2698 gfp_t gfp_mask) 2699 { 2700 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2701 } 2702 2703 2704 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2705 gfp_t gfp_mask) 2706 { 2707 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2708 } 2709 2710 2711 /** 2712 * skb_clone_writable - is the header of a clone writable 2713 * @skb: buffer to check 2714 * @len: length up to which to write 2715 * 2716 * Returns true if modifying the header part of the cloned buffer 2717 * does not requires the data to be copied. 2718 */ 2719 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2720 { 2721 return !skb_header_cloned(skb) && 2722 skb_headroom(skb) + len <= skb->hdr_len; 2723 } 2724 2725 static inline int skb_try_make_writable(struct sk_buff *skb, 2726 unsigned int write_len) 2727 { 2728 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2729 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2730 } 2731 2732 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2733 int cloned) 2734 { 2735 int delta = 0; 2736 2737 if (headroom > skb_headroom(skb)) 2738 delta = headroom - skb_headroom(skb); 2739 2740 if (delta || cloned) 2741 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2742 GFP_ATOMIC); 2743 return 0; 2744 } 2745 2746 /** 2747 * skb_cow - copy header of skb when it is required 2748 * @skb: buffer to cow 2749 * @headroom: needed headroom 2750 * 2751 * If the skb passed lacks sufficient headroom or its data part 2752 * is shared, data is reallocated. If reallocation fails, an error 2753 * is returned and original skb is not changed. 2754 * 2755 * The result is skb with writable area skb->head...skb->tail 2756 * and at least @headroom of space at head. 2757 */ 2758 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2759 { 2760 return __skb_cow(skb, headroom, skb_cloned(skb)); 2761 } 2762 2763 /** 2764 * skb_cow_head - skb_cow but only making the head writable 2765 * @skb: buffer to cow 2766 * @headroom: needed headroom 2767 * 2768 * This function is identical to skb_cow except that we replace the 2769 * skb_cloned check by skb_header_cloned. It should be used when 2770 * you only need to push on some header and do not need to modify 2771 * the data. 2772 */ 2773 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2774 { 2775 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2776 } 2777 2778 /** 2779 * skb_padto - pad an skbuff up to a minimal size 2780 * @skb: buffer to pad 2781 * @len: minimal length 2782 * 2783 * Pads up a buffer to ensure the trailing bytes exist and are 2784 * blanked. If the buffer already contains sufficient data it 2785 * is untouched. Otherwise it is extended. Returns zero on 2786 * success. The skb is freed on error. 2787 */ 2788 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2789 { 2790 unsigned int size = skb->len; 2791 if (likely(size >= len)) 2792 return 0; 2793 return skb_pad(skb, len - size); 2794 } 2795 2796 /** 2797 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2798 * @skb: buffer to pad 2799 * @len: minimal length 2800 * 2801 * Pads up a buffer to ensure the trailing bytes exist and are 2802 * blanked. If the buffer already contains sufficient data it 2803 * is untouched. Otherwise it is extended. Returns zero on 2804 * success. The skb is freed on error. 2805 */ 2806 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2807 { 2808 unsigned int size = skb->len; 2809 2810 if (unlikely(size < len)) { 2811 len -= size; 2812 if (skb_pad(skb, len)) 2813 return -ENOMEM; 2814 __skb_put(skb, len); 2815 } 2816 return 0; 2817 } 2818 2819 static inline int skb_add_data(struct sk_buff *skb, 2820 struct iov_iter *from, int copy) 2821 { 2822 const int off = skb->len; 2823 2824 if (skb->ip_summed == CHECKSUM_NONE) { 2825 __wsum csum = 0; 2826 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2827 &csum, from)) { 2828 skb->csum = csum_block_add(skb->csum, csum, off); 2829 return 0; 2830 } 2831 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 2832 return 0; 2833 2834 __skb_trim(skb, off); 2835 return -EFAULT; 2836 } 2837 2838 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2839 const struct page *page, int off) 2840 { 2841 if (i) { 2842 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2843 2844 return page == skb_frag_page(frag) && 2845 off == frag->page_offset + skb_frag_size(frag); 2846 } 2847 return false; 2848 } 2849 2850 static inline int __skb_linearize(struct sk_buff *skb) 2851 { 2852 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2853 } 2854 2855 /** 2856 * skb_linearize - convert paged skb to linear one 2857 * @skb: buffer to linarize 2858 * 2859 * If there is no free memory -ENOMEM is returned, otherwise zero 2860 * is returned and the old skb data released. 2861 */ 2862 static inline int skb_linearize(struct sk_buff *skb) 2863 { 2864 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2865 } 2866 2867 /** 2868 * skb_has_shared_frag - can any frag be overwritten 2869 * @skb: buffer to test 2870 * 2871 * Return true if the skb has at least one frag that might be modified 2872 * by an external entity (as in vmsplice()/sendfile()) 2873 */ 2874 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2875 { 2876 return skb_is_nonlinear(skb) && 2877 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2878 } 2879 2880 /** 2881 * skb_linearize_cow - make sure skb is linear and writable 2882 * @skb: buffer to process 2883 * 2884 * If there is no free memory -ENOMEM is returned, otherwise zero 2885 * is returned and the old skb data released. 2886 */ 2887 static inline int skb_linearize_cow(struct sk_buff *skb) 2888 { 2889 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2890 __skb_linearize(skb) : 0; 2891 } 2892 2893 static __always_inline void 2894 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 2895 unsigned int off) 2896 { 2897 if (skb->ip_summed == CHECKSUM_COMPLETE) 2898 skb->csum = csum_block_sub(skb->csum, 2899 csum_partial(start, len, 0), off); 2900 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2901 skb_checksum_start_offset(skb) < 0) 2902 skb->ip_summed = CHECKSUM_NONE; 2903 } 2904 2905 /** 2906 * skb_postpull_rcsum - update checksum for received skb after pull 2907 * @skb: buffer to update 2908 * @start: start of data before pull 2909 * @len: length of data pulled 2910 * 2911 * After doing a pull on a received packet, you need to call this to 2912 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2913 * CHECKSUM_NONE so that it can be recomputed from scratch. 2914 */ 2915 static inline void skb_postpull_rcsum(struct sk_buff *skb, 2916 const void *start, unsigned int len) 2917 { 2918 __skb_postpull_rcsum(skb, start, len, 0); 2919 } 2920 2921 static __always_inline void 2922 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 2923 unsigned int off) 2924 { 2925 if (skb->ip_summed == CHECKSUM_COMPLETE) 2926 skb->csum = csum_block_add(skb->csum, 2927 csum_partial(start, len, 0), off); 2928 } 2929 2930 /** 2931 * skb_postpush_rcsum - update checksum for received skb after push 2932 * @skb: buffer to update 2933 * @start: start of data after push 2934 * @len: length of data pushed 2935 * 2936 * After doing a push on a received packet, you need to call this to 2937 * update the CHECKSUM_COMPLETE checksum. 2938 */ 2939 static inline void skb_postpush_rcsum(struct sk_buff *skb, 2940 const void *start, unsigned int len) 2941 { 2942 __skb_postpush_rcsum(skb, start, len, 0); 2943 } 2944 2945 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2946 2947 /** 2948 * skb_push_rcsum - push skb and update receive checksum 2949 * @skb: buffer to update 2950 * @len: length of data pulled 2951 * 2952 * This function performs an skb_push on the packet and updates 2953 * the CHECKSUM_COMPLETE checksum. It should be used on 2954 * receive path processing instead of skb_push unless you know 2955 * that the checksum difference is zero (e.g., a valid IP header) 2956 * or you are setting ip_summed to CHECKSUM_NONE. 2957 */ 2958 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb, 2959 unsigned int len) 2960 { 2961 skb_push(skb, len); 2962 skb_postpush_rcsum(skb, skb->data, len); 2963 return skb->data; 2964 } 2965 2966 /** 2967 * pskb_trim_rcsum - trim received skb and update checksum 2968 * @skb: buffer to trim 2969 * @len: new length 2970 * 2971 * This is exactly the same as pskb_trim except that it ensures the 2972 * checksum of received packets are still valid after the operation. 2973 */ 2974 2975 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2976 { 2977 if (likely(len >= skb->len)) 2978 return 0; 2979 if (skb->ip_summed == CHECKSUM_COMPLETE) 2980 skb->ip_summed = CHECKSUM_NONE; 2981 return __pskb_trim(skb, len); 2982 } 2983 2984 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2985 { 2986 if (skb->ip_summed == CHECKSUM_COMPLETE) 2987 skb->ip_summed = CHECKSUM_NONE; 2988 __skb_trim(skb, len); 2989 return 0; 2990 } 2991 2992 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 2993 { 2994 if (skb->ip_summed == CHECKSUM_COMPLETE) 2995 skb->ip_summed = CHECKSUM_NONE; 2996 return __skb_grow(skb, len); 2997 } 2998 2999 #define skb_queue_walk(queue, skb) \ 3000 for (skb = (queue)->next; \ 3001 skb != (struct sk_buff *)(queue); \ 3002 skb = skb->next) 3003 3004 #define skb_queue_walk_safe(queue, skb, tmp) \ 3005 for (skb = (queue)->next, tmp = skb->next; \ 3006 skb != (struct sk_buff *)(queue); \ 3007 skb = tmp, tmp = skb->next) 3008 3009 #define skb_queue_walk_from(queue, skb) \ 3010 for (; skb != (struct sk_buff *)(queue); \ 3011 skb = skb->next) 3012 3013 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3014 for (tmp = skb->next; \ 3015 skb != (struct sk_buff *)(queue); \ 3016 skb = tmp, tmp = skb->next) 3017 3018 #define skb_queue_reverse_walk(queue, skb) \ 3019 for (skb = (queue)->prev; \ 3020 skb != (struct sk_buff *)(queue); \ 3021 skb = skb->prev) 3022 3023 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3024 for (skb = (queue)->prev, tmp = skb->prev; \ 3025 skb != (struct sk_buff *)(queue); \ 3026 skb = tmp, tmp = skb->prev) 3027 3028 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3029 for (tmp = skb->prev; \ 3030 skb != (struct sk_buff *)(queue); \ 3031 skb = tmp, tmp = skb->prev) 3032 3033 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3034 { 3035 return skb_shinfo(skb)->frag_list != NULL; 3036 } 3037 3038 static inline void skb_frag_list_init(struct sk_buff *skb) 3039 { 3040 skb_shinfo(skb)->frag_list = NULL; 3041 } 3042 3043 #define skb_walk_frags(skb, iter) \ 3044 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3045 3046 3047 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3048 const struct sk_buff *skb); 3049 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3050 void (*destructor)(struct sock *sk, 3051 struct sk_buff *skb), 3052 int *peeked, int *off, int *err, 3053 struct sk_buff **last); 3054 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3055 void (*destructor)(struct sock *sk, 3056 struct sk_buff *skb), 3057 int *peeked, int *off, int *err); 3058 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3059 int *err); 3060 unsigned int datagram_poll(struct file *file, struct socket *sock, 3061 struct poll_table_struct *wait); 3062 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3063 struct iov_iter *to, int size); 3064 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3065 struct msghdr *msg, int size) 3066 { 3067 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3068 } 3069 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3070 struct msghdr *msg); 3071 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3072 struct iov_iter *from, int len); 3073 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3074 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3075 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3076 static inline void skb_free_datagram_locked(struct sock *sk, 3077 struct sk_buff *skb) 3078 { 3079 __skb_free_datagram_locked(sk, skb, 0); 3080 } 3081 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3082 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3083 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3084 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3085 int len, __wsum csum); 3086 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3087 struct pipe_inode_info *pipe, unsigned int len, 3088 unsigned int flags); 3089 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3090 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3091 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3092 int len, int hlen); 3093 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3094 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3095 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3096 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3097 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu); 3098 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3099 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3100 int skb_ensure_writable(struct sk_buff *skb, int write_len); 3101 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3102 int skb_vlan_pop(struct sk_buff *skb); 3103 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3104 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3105 gfp_t gfp); 3106 3107 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3108 { 3109 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3110 } 3111 3112 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3113 { 3114 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3115 } 3116 3117 struct skb_checksum_ops { 3118 __wsum (*update)(const void *mem, int len, __wsum wsum); 3119 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3120 }; 3121 3122 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3123 __wsum csum, const struct skb_checksum_ops *ops); 3124 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3125 __wsum csum); 3126 3127 static inline void * __must_check 3128 __skb_header_pointer(const struct sk_buff *skb, int offset, 3129 int len, void *data, int hlen, void *buffer) 3130 { 3131 if (hlen - offset >= len) 3132 return data + offset; 3133 3134 if (!skb || 3135 skb_copy_bits(skb, offset, buffer, len) < 0) 3136 return NULL; 3137 3138 return buffer; 3139 } 3140 3141 static inline void * __must_check 3142 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3143 { 3144 return __skb_header_pointer(skb, offset, len, skb->data, 3145 skb_headlen(skb), buffer); 3146 } 3147 3148 /** 3149 * skb_needs_linearize - check if we need to linearize a given skb 3150 * depending on the given device features. 3151 * @skb: socket buffer to check 3152 * @features: net device features 3153 * 3154 * Returns true if either: 3155 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3156 * 2. skb is fragmented and the device does not support SG. 3157 */ 3158 static inline bool skb_needs_linearize(struct sk_buff *skb, 3159 netdev_features_t features) 3160 { 3161 return skb_is_nonlinear(skb) && 3162 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3163 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3164 } 3165 3166 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3167 void *to, 3168 const unsigned int len) 3169 { 3170 memcpy(to, skb->data, len); 3171 } 3172 3173 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3174 const int offset, void *to, 3175 const unsigned int len) 3176 { 3177 memcpy(to, skb->data + offset, len); 3178 } 3179 3180 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3181 const void *from, 3182 const unsigned int len) 3183 { 3184 memcpy(skb->data, from, len); 3185 } 3186 3187 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3188 const int offset, 3189 const void *from, 3190 const unsigned int len) 3191 { 3192 memcpy(skb->data + offset, from, len); 3193 } 3194 3195 void skb_init(void); 3196 3197 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3198 { 3199 return skb->tstamp; 3200 } 3201 3202 /** 3203 * skb_get_timestamp - get timestamp from a skb 3204 * @skb: skb to get stamp from 3205 * @stamp: pointer to struct timeval to store stamp in 3206 * 3207 * Timestamps are stored in the skb as offsets to a base timestamp. 3208 * This function converts the offset back to a struct timeval and stores 3209 * it in stamp. 3210 */ 3211 static inline void skb_get_timestamp(const struct sk_buff *skb, 3212 struct timeval *stamp) 3213 { 3214 *stamp = ktime_to_timeval(skb->tstamp); 3215 } 3216 3217 static inline void skb_get_timestampns(const struct sk_buff *skb, 3218 struct timespec *stamp) 3219 { 3220 *stamp = ktime_to_timespec(skb->tstamp); 3221 } 3222 3223 static inline void __net_timestamp(struct sk_buff *skb) 3224 { 3225 skb->tstamp = ktime_get_real(); 3226 } 3227 3228 static inline ktime_t net_timedelta(ktime_t t) 3229 { 3230 return ktime_sub(ktime_get_real(), t); 3231 } 3232 3233 static inline ktime_t net_invalid_timestamp(void) 3234 { 3235 return 0; 3236 } 3237 3238 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3239 3240 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3241 3242 void skb_clone_tx_timestamp(struct sk_buff *skb); 3243 bool skb_defer_rx_timestamp(struct sk_buff *skb); 3244 3245 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3246 3247 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3248 { 3249 } 3250 3251 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3252 { 3253 return false; 3254 } 3255 3256 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3257 3258 /** 3259 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3260 * 3261 * PHY drivers may accept clones of transmitted packets for 3262 * timestamping via their phy_driver.txtstamp method. These drivers 3263 * must call this function to return the skb back to the stack with a 3264 * timestamp. 3265 * 3266 * @skb: clone of the the original outgoing packet 3267 * @hwtstamps: hardware time stamps 3268 * 3269 */ 3270 void skb_complete_tx_timestamp(struct sk_buff *skb, 3271 struct skb_shared_hwtstamps *hwtstamps); 3272 3273 void __skb_tstamp_tx(struct sk_buff *orig_skb, 3274 struct skb_shared_hwtstamps *hwtstamps, 3275 struct sock *sk, int tstype); 3276 3277 /** 3278 * skb_tstamp_tx - queue clone of skb with send time stamps 3279 * @orig_skb: the original outgoing packet 3280 * @hwtstamps: hardware time stamps, may be NULL if not available 3281 * 3282 * If the skb has a socket associated, then this function clones the 3283 * skb (thus sharing the actual data and optional structures), stores 3284 * the optional hardware time stamping information (if non NULL) or 3285 * generates a software time stamp (otherwise), then queues the clone 3286 * to the error queue of the socket. Errors are silently ignored. 3287 */ 3288 void skb_tstamp_tx(struct sk_buff *orig_skb, 3289 struct skb_shared_hwtstamps *hwtstamps); 3290 3291 static inline void sw_tx_timestamp(struct sk_buff *skb) 3292 { 3293 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3294 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3295 skb_tstamp_tx(skb, NULL); 3296 } 3297 3298 /** 3299 * skb_tx_timestamp() - Driver hook for transmit timestamping 3300 * 3301 * Ethernet MAC Drivers should call this function in their hard_xmit() 3302 * function immediately before giving the sk_buff to the MAC hardware. 3303 * 3304 * Specifically, one should make absolutely sure that this function is 3305 * called before TX completion of this packet can trigger. Otherwise 3306 * the packet could potentially already be freed. 3307 * 3308 * @skb: A socket buffer. 3309 */ 3310 static inline void skb_tx_timestamp(struct sk_buff *skb) 3311 { 3312 skb_clone_tx_timestamp(skb); 3313 sw_tx_timestamp(skb); 3314 } 3315 3316 /** 3317 * skb_complete_wifi_ack - deliver skb with wifi status 3318 * 3319 * @skb: the original outgoing packet 3320 * @acked: ack status 3321 * 3322 */ 3323 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3324 3325 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3326 __sum16 __skb_checksum_complete(struct sk_buff *skb); 3327 3328 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3329 { 3330 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3331 skb->csum_valid || 3332 (skb->ip_summed == CHECKSUM_PARTIAL && 3333 skb_checksum_start_offset(skb) >= 0)); 3334 } 3335 3336 /** 3337 * skb_checksum_complete - Calculate checksum of an entire packet 3338 * @skb: packet to process 3339 * 3340 * This function calculates the checksum over the entire packet plus 3341 * the value of skb->csum. The latter can be used to supply the 3342 * checksum of a pseudo header as used by TCP/UDP. It returns the 3343 * checksum. 3344 * 3345 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3346 * this function can be used to verify that checksum on received 3347 * packets. In that case the function should return zero if the 3348 * checksum is correct. In particular, this function will return zero 3349 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3350 * hardware has already verified the correctness of the checksum. 3351 */ 3352 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3353 { 3354 return skb_csum_unnecessary(skb) ? 3355 0 : __skb_checksum_complete(skb); 3356 } 3357 3358 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3359 { 3360 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3361 if (skb->csum_level == 0) 3362 skb->ip_summed = CHECKSUM_NONE; 3363 else 3364 skb->csum_level--; 3365 } 3366 } 3367 3368 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3369 { 3370 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3371 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3372 skb->csum_level++; 3373 } else if (skb->ip_summed == CHECKSUM_NONE) { 3374 skb->ip_summed = CHECKSUM_UNNECESSARY; 3375 skb->csum_level = 0; 3376 } 3377 } 3378 3379 static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3380 { 3381 /* Mark current checksum as bad (typically called from GRO 3382 * path). In the case that ip_summed is CHECKSUM_NONE 3383 * this must be the first checksum encountered in the packet. 3384 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3385 * checksum after the last one validated. For UDP, a zero 3386 * checksum can not be marked as bad. 3387 */ 3388 3389 if (skb->ip_summed == CHECKSUM_NONE || 3390 skb->ip_summed == CHECKSUM_UNNECESSARY) 3391 skb->csum_bad = 1; 3392 } 3393 3394 /* Check if we need to perform checksum complete validation. 3395 * 3396 * Returns true if checksum complete is needed, false otherwise 3397 * (either checksum is unnecessary or zero checksum is allowed). 3398 */ 3399 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3400 bool zero_okay, 3401 __sum16 check) 3402 { 3403 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3404 skb->csum_valid = 1; 3405 __skb_decr_checksum_unnecessary(skb); 3406 return false; 3407 } 3408 3409 return true; 3410 } 3411 3412 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3413 * in checksum_init. 3414 */ 3415 #define CHECKSUM_BREAK 76 3416 3417 /* Unset checksum-complete 3418 * 3419 * Unset checksum complete can be done when packet is being modified 3420 * (uncompressed for instance) and checksum-complete value is 3421 * invalidated. 3422 */ 3423 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3424 { 3425 if (skb->ip_summed == CHECKSUM_COMPLETE) 3426 skb->ip_summed = CHECKSUM_NONE; 3427 } 3428 3429 /* Validate (init) checksum based on checksum complete. 3430 * 3431 * Return values: 3432 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3433 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3434 * checksum is stored in skb->csum for use in __skb_checksum_complete 3435 * non-zero: value of invalid checksum 3436 * 3437 */ 3438 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3439 bool complete, 3440 __wsum psum) 3441 { 3442 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3443 if (!csum_fold(csum_add(psum, skb->csum))) { 3444 skb->csum_valid = 1; 3445 return 0; 3446 } 3447 } else if (skb->csum_bad) { 3448 /* ip_summed == CHECKSUM_NONE in this case */ 3449 return (__force __sum16)1; 3450 } 3451 3452 skb->csum = psum; 3453 3454 if (complete || skb->len <= CHECKSUM_BREAK) { 3455 __sum16 csum; 3456 3457 csum = __skb_checksum_complete(skb); 3458 skb->csum_valid = !csum; 3459 return csum; 3460 } 3461 3462 return 0; 3463 } 3464 3465 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3466 { 3467 return 0; 3468 } 3469 3470 /* Perform checksum validate (init). Note that this is a macro since we only 3471 * want to calculate the pseudo header which is an input function if necessary. 3472 * First we try to validate without any computation (checksum unnecessary) and 3473 * then calculate based on checksum complete calling the function to compute 3474 * pseudo header. 3475 * 3476 * Return values: 3477 * 0: checksum is validated or try to in skb_checksum_complete 3478 * non-zero: value of invalid checksum 3479 */ 3480 #define __skb_checksum_validate(skb, proto, complete, \ 3481 zero_okay, check, compute_pseudo) \ 3482 ({ \ 3483 __sum16 __ret = 0; \ 3484 skb->csum_valid = 0; \ 3485 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3486 __ret = __skb_checksum_validate_complete(skb, \ 3487 complete, compute_pseudo(skb, proto)); \ 3488 __ret; \ 3489 }) 3490 3491 #define skb_checksum_init(skb, proto, compute_pseudo) \ 3492 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3493 3494 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3495 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3496 3497 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 3498 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3499 3500 #define skb_checksum_validate_zero_check(skb, proto, check, \ 3501 compute_pseudo) \ 3502 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3503 3504 #define skb_checksum_simple_validate(skb) \ 3505 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3506 3507 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3508 { 3509 return (skb->ip_summed == CHECKSUM_NONE && 3510 skb->csum_valid && !skb->csum_bad); 3511 } 3512 3513 static inline void __skb_checksum_convert(struct sk_buff *skb, 3514 __sum16 check, __wsum pseudo) 3515 { 3516 skb->csum = ~pseudo; 3517 skb->ip_summed = CHECKSUM_COMPLETE; 3518 } 3519 3520 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3521 do { \ 3522 if (__skb_checksum_convert_check(skb)) \ 3523 __skb_checksum_convert(skb, check, \ 3524 compute_pseudo(skb, proto)); \ 3525 } while (0) 3526 3527 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3528 u16 start, u16 offset) 3529 { 3530 skb->ip_summed = CHECKSUM_PARTIAL; 3531 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3532 skb->csum_offset = offset - start; 3533 } 3534 3535 /* Update skbuf and packet to reflect the remote checksum offload operation. 3536 * When called, ptr indicates the starting point for skb->csum when 3537 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3538 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3539 */ 3540 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3541 int start, int offset, bool nopartial) 3542 { 3543 __wsum delta; 3544 3545 if (!nopartial) { 3546 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3547 return; 3548 } 3549 3550 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3551 __skb_checksum_complete(skb); 3552 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3553 } 3554 3555 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3556 3557 /* Adjust skb->csum since we changed the packet */ 3558 skb->csum = csum_add(skb->csum, delta); 3559 } 3560 3561 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3562 void nf_conntrack_destroy(struct nf_conntrack *nfct); 3563 static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3564 { 3565 if (nfct && atomic_dec_and_test(&nfct->use)) 3566 nf_conntrack_destroy(nfct); 3567 } 3568 static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3569 { 3570 if (nfct) 3571 atomic_inc(&nfct->use); 3572 } 3573 #endif 3574 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3575 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3576 { 3577 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3578 kfree(nf_bridge); 3579 } 3580 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3581 { 3582 if (nf_bridge) 3583 atomic_inc(&nf_bridge->use); 3584 } 3585 #endif /* CONFIG_BRIDGE_NETFILTER */ 3586 static inline void nf_reset(struct sk_buff *skb) 3587 { 3588 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3589 nf_conntrack_put(skb->nfct); 3590 skb->nfct = NULL; 3591 #endif 3592 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3593 nf_bridge_put(skb->nf_bridge); 3594 skb->nf_bridge = NULL; 3595 #endif 3596 } 3597 3598 static inline void nf_reset_trace(struct sk_buff *skb) 3599 { 3600 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3601 skb->nf_trace = 0; 3602 #endif 3603 } 3604 3605 /* Note: This doesn't put any conntrack and bridge info in dst. */ 3606 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3607 bool copy) 3608 { 3609 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3610 dst->nfct = src->nfct; 3611 nf_conntrack_get(src->nfct); 3612 if (copy) 3613 dst->nfctinfo = src->nfctinfo; 3614 #endif 3615 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3616 dst->nf_bridge = src->nf_bridge; 3617 nf_bridge_get(src->nf_bridge); 3618 #endif 3619 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3620 if (copy) 3621 dst->nf_trace = src->nf_trace; 3622 #endif 3623 } 3624 3625 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3626 { 3627 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3628 nf_conntrack_put(dst->nfct); 3629 #endif 3630 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3631 nf_bridge_put(dst->nf_bridge); 3632 #endif 3633 __nf_copy(dst, src, true); 3634 } 3635 3636 #ifdef CONFIG_NETWORK_SECMARK 3637 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3638 { 3639 to->secmark = from->secmark; 3640 } 3641 3642 static inline void skb_init_secmark(struct sk_buff *skb) 3643 { 3644 skb->secmark = 0; 3645 } 3646 #else 3647 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3648 { } 3649 3650 static inline void skb_init_secmark(struct sk_buff *skb) 3651 { } 3652 #endif 3653 3654 static inline bool skb_irq_freeable(const struct sk_buff *skb) 3655 { 3656 return !skb->destructor && 3657 #if IS_ENABLED(CONFIG_XFRM) 3658 !skb->sp && 3659 #endif 3660 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 3661 !skb->nfct && 3662 #endif 3663 !skb->_skb_refdst && 3664 !skb_has_frag_list(skb); 3665 } 3666 3667 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3668 { 3669 skb->queue_mapping = queue_mapping; 3670 } 3671 3672 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3673 { 3674 return skb->queue_mapping; 3675 } 3676 3677 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3678 { 3679 to->queue_mapping = from->queue_mapping; 3680 } 3681 3682 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3683 { 3684 skb->queue_mapping = rx_queue + 1; 3685 } 3686 3687 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3688 { 3689 return skb->queue_mapping - 1; 3690 } 3691 3692 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3693 { 3694 return skb->queue_mapping != 0; 3695 } 3696 3697 static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3698 { 3699 #ifdef CONFIG_XFRM 3700 return skb->sp; 3701 #else 3702 return NULL; 3703 #endif 3704 } 3705 3706 /* Keeps track of mac header offset relative to skb->head. 3707 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3708 * For non-tunnel skb it points to skb_mac_header() and for 3709 * tunnel skb it points to outer mac header. 3710 * Keeps track of level of encapsulation of network headers. 3711 */ 3712 struct skb_gso_cb { 3713 union { 3714 int mac_offset; 3715 int data_offset; 3716 }; 3717 int encap_level; 3718 __wsum csum; 3719 __u16 csum_start; 3720 }; 3721 #define SKB_SGO_CB_OFFSET 32 3722 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3723 3724 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3725 { 3726 return (skb_mac_header(inner_skb) - inner_skb->head) - 3727 SKB_GSO_CB(inner_skb)->mac_offset; 3728 } 3729 3730 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3731 { 3732 int new_headroom, headroom; 3733 int ret; 3734 3735 headroom = skb_headroom(skb); 3736 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3737 if (ret) 3738 return ret; 3739 3740 new_headroom = skb_headroom(skb); 3741 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3742 return 0; 3743 } 3744 3745 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3746 { 3747 /* Do not update partial checksums if remote checksum is enabled. */ 3748 if (skb->remcsum_offload) 3749 return; 3750 3751 SKB_GSO_CB(skb)->csum = res; 3752 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3753 } 3754 3755 /* Compute the checksum for a gso segment. First compute the checksum value 3756 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3757 * then add in skb->csum (checksum from csum_start to end of packet). 3758 * skb->csum and csum_start are then updated to reflect the checksum of the 3759 * resultant packet starting from the transport header-- the resultant checksum 3760 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3761 * header. 3762 */ 3763 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3764 { 3765 unsigned char *csum_start = skb_transport_header(skb); 3766 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3767 __wsum partial = SKB_GSO_CB(skb)->csum; 3768 3769 SKB_GSO_CB(skb)->csum = res; 3770 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3771 3772 return csum_fold(csum_partial(csum_start, plen, partial)); 3773 } 3774 3775 static inline bool skb_is_gso(const struct sk_buff *skb) 3776 { 3777 return skb_shinfo(skb)->gso_size; 3778 } 3779 3780 /* Note: Should be called only if skb_is_gso(skb) is true */ 3781 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3782 { 3783 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3784 } 3785 3786 static inline void skb_gso_reset(struct sk_buff *skb) 3787 { 3788 skb_shinfo(skb)->gso_size = 0; 3789 skb_shinfo(skb)->gso_segs = 0; 3790 skb_shinfo(skb)->gso_type = 0; 3791 } 3792 3793 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3794 3795 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3796 { 3797 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3798 * wanted then gso_type will be set. */ 3799 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3800 3801 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3802 unlikely(shinfo->gso_type == 0)) { 3803 __skb_warn_lro_forwarding(skb); 3804 return true; 3805 } 3806 return false; 3807 } 3808 3809 static inline void skb_forward_csum(struct sk_buff *skb) 3810 { 3811 /* Unfortunately we don't support this one. Any brave souls? */ 3812 if (skb->ip_summed == CHECKSUM_COMPLETE) 3813 skb->ip_summed = CHECKSUM_NONE; 3814 } 3815 3816 /** 3817 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3818 * @skb: skb to check 3819 * 3820 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3821 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3822 * use this helper, to document places where we make this assertion. 3823 */ 3824 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3825 { 3826 #ifdef DEBUG 3827 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3828 #endif 3829 } 3830 3831 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3832 3833 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3834 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3835 unsigned int transport_len, 3836 __sum16(*skb_chkf)(struct sk_buff *skb)); 3837 3838 /** 3839 * skb_head_is_locked - Determine if the skb->head is locked down 3840 * @skb: skb to check 3841 * 3842 * The head on skbs build around a head frag can be removed if they are 3843 * not cloned. This function returns true if the skb head is locked down 3844 * due to either being allocated via kmalloc, or by being a clone with 3845 * multiple references to the head. 3846 */ 3847 static inline bool skb_head_is_locked(const struct sk_buff *skb) 3848 { 3849 return !skb->head_frag || skb_cloned(skb); 3850 } 3851 3852 /** 3853 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3854 * 3855 * @skb: GSO skb 3856 * 3857 * skb_gso_network_seglen is used to determine the real size of the 3858 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3859 * 3860 * The MAC/L2 header is not accounted for. 3861 */ 3862 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3863 { 3864 unsigned int hdr_len = skb_transport_header(skb) - 3865 skb_network_header(skb); 3866 return hdr_len + skb_gso_transport_seglen(skb); 3867 } 3868 3869 /* Local Checksum Offload. 3870 * Compute outer checksum based on the assumption that the 3871 * inner checksum will be offloaded later. 3872 * See Documentation/networking/checksum-offloads.txt for 3873 * explanation of how this works. 3874 * Fill in outer checksum adjustment (e.g. with sum of outer 3875 * pseudo-header) before calling. 3876 * Also ensure that inner checksum is in linear data area. 3877 */ 3878 static inline __wsum lco_csum(struct sk_buff *skb) 3879 { 3880 unsigned char *csum_start = skb_checksum_start(skb); 3881 unsigned char *l4_hdr = skb_transport_header(skb); 3882 __wsum partial; 3883 3884 /* Start with complement of inner checksum adjustment */ 3885 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 3886 skb->csum_offset)); 3887 3888 /* Add in checksum of our headers (incl. outer checksum 3889 * adjustment filled in by caller) and return result. 3890 */ 3891 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 3892 } 3893 3894 #endif /* __KERNEL__ */ 3895 #endif /* _LINUX_SKBUFF_H */ 3896